Add files via upload

2020-10-21 15:25:32 -04:00 · 2020-10-21 15:25:32 -04:00 · 4db12ac10b
--- a/README.md
+++ b/README.md
@ -1,14 +1,66 @@
-# Project
+## DIGITS EXPERIMENTS

-> This repo has been populated by an initial template to help get you started. Please
-> make sure to update the content to build a great experience for community-building.
+To download the datasets, you can e.g. unpack the following [file](https://github.com/thuml/CDAN/blob/master/data/usps2mnist/images.tar.gz) in data/digits.
+For the sake of speed when running experiments, the code generates a pkl file containing the whole datasets (when first run) and then loads it at runtime.

-As the maintainer of this project, please make a few updates:
+To run the code on the original datasets, run:

- Improving this README.MD file to provide a great experience
- Updating SUPPORT.MD with content about this project's support experience
- Understanding the security reporting process in SECURITY.MD
- Remove this section from the README
+`python train_digits.py {method} --task {task}`
+
+where method belongs to: [`CDAN`, `CDAN-E`, `DANN`, `IWDAN`, `NANN`, `IWDANORACLE`, `IWCDAN`, `IWCDANORACLE`, `IWCDAN-E`, `IWCDAN-EORACLE`] and task is either mnist2usps or usps2mnist.
+
+To run the code on the subsampled datasets, run:
+
+`python train_digits.py {method} --task {task} --ratio 1`
+
+To reproduce Figs. 1 and 2, run the following command with various seeds:
+
+`python train_digits.py {method} --ratio {ratio}`
+
+where ratio belongs to 100 <= ratio < 150 (to subsample the target) or to 200 <= ratio < 250 (to subsample the source). Each corresponds to a given subsampling, the exact fractions can be found in the subsampling list of `data_list.txt.
+
+## VISDA AND OFFICE DATASETS
+
+The Visda dataset can be found here: https://github.com/VisionLearningGroup/taskcv-2017-public.
+
+The Office-31 dataset can be found here: https://people.eecs.berkeley.edu/~jhoffman/domainadapt.
+
+The Office-Home dataset can be found here: http://hemanthdv.org/OfficeHome-Dataset.
+
+They should be downloaded and placed in the corresponding folder under data. The code will generate a test file once for faster evaluation, it might take a while during the first visda run.
+
+### Discriminator based methods
+
+To run the code on the original datasets, run:
+
+`python train_image.py {method} --dset {dset} --s_dset_file {s_dset_file} --t_dset_file {t_dset_file}`
+
+where:
+  - `method` belongs to [`CDAN`, `CDAN-E`, `DANN`, `IWDAN`, `NANN`, `IWDANORACLE`, `IWCDAN`, `IWCDANORACLE`, `IWCDAN-E`, `IWCDAN-EORACLE`]
+  - `dset` belongs to [`visda`, `office-31`, `office-home`]
+  - `s_dset_file` corresponds to the source domain, the filename can be found in the corresponding data folder, e.g. `dslr_list.txt` (not needed for VISDA)
+  - `t_dset_file` corresponds to the target domain, the filename can be found in the corresponding data folder, e.g. `amazon_list.txt` (not needed for VISDA).
+
+To run the code on the subsampled datasets, run the same command with `--ratio 1` appended to it.
+
+### MMD-based methods
+
+To run the MMD algorithms (e.g. IWJAN), use the same commands as above with the `train_mmd.py` file.
+
+## Reference
+
+Please use the following bibtex entry if you use this code for SPIBB:
+
+```
+@misc{tachet2020domain,
+      title={Domain Adaptation with Conditional Distribution Matching and Generalized Label Shift},
+      author={Tachet des Combes, Remi and Zhao, Han and Wang, Yu-Xiang and Gordon, Geoff},
+      year={2020},
+      eprint={2003.04475},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG}
+}
+```

 ## Contributing

@ -26,8 +78,8 @@ contact [opencode@microsoft.com](mailto:opencode@microsoft.com) with any additio

 ## Trademarks

-This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft 
-trademarks or logos is subject to and must follow 
+This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft
+trademarks or logos is subject to and must follow
 [Microsoft's Trademark & Brand Guidelines](https://www.microsoft.com/en-us/legal/intellectualproperty/trademarks/usage/general).
 Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship.
-Any use of third-party trademarks or logos are subject to those third-party's policies.
+Any use of third-party trademarks or logos are subject to those third-party's policies.
--- a/data/office-31/amazon_list.txt
+++ b/data/office-31/amazon_list.txt
--- a/data/office-31/dslr_list.txt
+++ b/data/office-31/dslr_list.txt
@ -0,0 +1,498 @@
+dslr/images/calculator/frame_0001.jpg 5
+dslr/images/calculator/frame_0002.jpg 5
+dslr/images/calculator/frame_0003.jpg 5
+dslr/images/calculator/frame_0004.jpg 5
+dslr/images/calculator/frame_0005.jpg 5
+dslr/images/calculator/frame_0006.jpg 5
+dslr/images/calculator/frame_0007.jpg 5
+dslr/images/calculator/frame_0008.jpg 5
+dslr/images/calculator/frame_0009.jpg 5
+dslr/images/calculator/frame_0010.jpg 5
+dslr/images/calculator/frame_0011.jpg 5
+dslr/images/calculator/frame_0012.jpg 5
+dslr/images/ring_binder/frame_0001.jpg 24
+dslr/images/ring_binder/frame_0002.jpg 24
+dslr/images/ring_binder/frame_0003.jpg 24
+dslr/images/ring_binder/frame_0004.jpg 24
+dslr/images/ring_binder/frame_0005.jpg 24
+dslr/images/ring_binder/frame_0006.jpg 24
+dslr/images/ring_binder/frame_0007.jpg 24
+dslr/images/ring_binder/frame_0008.jpg 24
+dslr/images/ring_binder/frame_0009.jpg 24
+dslr/images/ring_binder/frame_0010.jpg 24
+dslr/images/printer/frame_0001.jpg 21
+dslr/images/printer/frame_0002.jpg 21
+dslr/images/printer/frame_0003.jpg 21
+dslr/images/printer/frame_0004.jpg 21
+dslr/images/printer/frame_0005.jpg 21
+dslr/images/printer/frame_0006.jpg 21
+dslr/images/printer/frame_0007.jpg 21
+dslr/images/printer/frame_0008.jpg 21
+dslr/images/printer/frame_0009.jpg 21
+dslr/images/printer/frame_0010.jpg 21
+dslr/images/printer/frame_0011.jpg 21
+dslr/images/printer/frame_0012.jpg 21
+dslr/images/printer/frame_0013.jpg 21
+dslr/images/printer/frame_0014.jpg 21
+dslr/images/printer/frame_0015.jpg 21
+dslr/images/keyboard/frame_0001.jpg 11
+dslr/images/keyboard/frame_0002.jpg 11
+dslr/images/keyboard/frame_0003.jpg 11
+dslr/images/keyboard/frame_0004.jpg 11
+dslr/images/keyboard/frame_0005.jpg 11
+dslr/images/keyboard/frame_0006.jpg 11
+dslr/images/keyboard/frame_0007.jpg 11
+dslr/images/keyboard/frame_0008.jpg 11
+dslr/images/keyboard/frame_0009.jpg 11
+dslr/images/keyboard/frame_0010.jpg 11
+dslr/images/scissors/frame_0001.jpg 26
+dslr/images/scissors/frame_0002.jpg 26
+dslr/images/scissors/frame_0003.jpg 26
+dslr/images/scissors/frame_0004.jpg 26
+dslr/images/scissors/frame_0005.jpg 26
+dslr/images/scissors/frame_0006.jpg 26
+dslr/images/scissors/frame_0007.jpg 26
+dslr/images/scissors/frame_0008.jpg 26
+dslr/images/scissors/frame_0009.jpg 26
+dslr/images/scissors/frame_0010.jpg 26
+dslr/images/scissors/frame_0011.jpg 26
+dslr/images/scissors/frame_0012.jpg 26
+dslr/images/scissors/frame_0013.jpg 26
+dslr/images/scissors/frame_0014.jpg 26
+dslr/images/scissors/frame_0015.jpg 26
+dslr/images/scissors/frame_0016.jpg 26
+dslr/images/scissors/frame_0017.jpg 26
+dslr/images/scissors/frame_0018.jpg 26
+dslr/images/laptop_computer/frame_0001.jpg 12
+dslr/images/laptop_computer/frame_0002.jpg 12
+dslr/images/laptop_computer/frame_0003.jpg 12
+dslr/images/laptop_computer/frame_0004.jpg 12
+dslr/images/laptop_computer/frame_0005.jpg 12
+dslr/images/laptop_computer/frame_0006.jpg 12
+dslr/images/laptop_computer/frame_0007.jpg 12
+dslr/images/laptop_computer/frame_0008.jpg 12
+dslr/images/laptop_computer/frame_0009.jpg 12
+dslr/images/laptop_computer/frame_0010.jpg 12
+dslr/images/laptop_computer/frame_0011.jpg 12
+dslr/images/laptop_computer/frame_0012.jpg 12
+dslr/images/laptop_computer/frame_0013.jpg 12
+dslr/images/laptop_computer/frame_0014.jpg 12
+dslr/images/laptop_computer/frame_0015.jpg 12
+dslr/images/laptop_computer/frame_0016.jpg 12
+dslr/images/laptop_computer/frame_0017.jpg 12
+dslr/images/laptop_computer/frame_0018.jpg 12
+dslr/images/laptop_computer/frame_0019.jpg 12
+dslr/images/laptop_computer/frame_0020.jpg 12
+dslr/images/laptop_computer/frame_0021.jpg 12
+dslr/images/laptop_computer/frame_0022.jpg 12
+dslr/images/laptop_computer/frame_0023.jpg 12
+dslr/images/laptop_computer/frame_0024.jpg 12
+dslr/images/mouse/frame_0001.jpg 16
+dslr/images/mouse/frame_0002.jpg 16
+dslr/images/mouse/frame_0003.jpg 16
+dslr/images/mouse/frame_0004.jpg 16
+dslr/images/mouse/frame_0005.jpg 16
+dslr/images/mouse/frame_0006.jpg 16
+dslr/images/mouse/frame_0007.jpg 16
+dslr/images/mouse/frame_0008.jpg 16
+dslr/images/mouse/frame_0009.jpg 16
+dslr/images/mouse/frame_0010.jpg 16
+dslr/images/mouse/frame_0011.jpg 16
+dslr/images/mouse/frame_0012.jpg 16
+dslr/images/monitor/frame_0001.jpg 15
+dslr/images/monitor/frame_0002.jpg 15
+dslr/images/monitor/frame_0003.jpg 15
+dslr/images/monitor/frame_0004.jpg 15
+dslr/images/monitor/frame_0005.jpg 15
+dslr/images/monitor/frame_0006.jpg 15
+dslr/images/monitor/frame_0007.jpg 15
+dslr/images/monitor/frame_0008.jpg 15
+dslr/images/monitor/frame_0009.jpg 15
+dslr/images/monitor/frame_0010.jpg 15
+dslr/images/monitor/frame_0011.jpg 15
+dslr/images/monitor/frame_0012.jpg 15
+dslr/images/monitor/frame_0013.jpg 15
+dslr/images/monitor/frame_0014.jpg 15
+dslr/images/monitor/frame_0015.jpg 15
+dslr/images/monitor/frame_0016.jpg 15
+dslr/images/monitor/frame_0017.jpg 15
+dslr/images/monitor/frame_0018.jpg 15
+dslr/images/monitor/frame_0019.jpg 15
+dslr/images/monitor/frame_0020.jpg 15
+dslr/images/monitor/frame_0021.jpg 15
+dslr/images/monitor/frame_0022.jpg 15
+dslr/images/mug/frame_0001.jpg 17
+dslr/images/mug/frame_0002.jpg 17
+dslr/images/mug/frame_0003.jpg 17
+dslr/images/mug/frame_0004.jpg 17
+dslr/images/mug/frame_0005.jpg 17
+dslr/images/mug/frame_0006.jpg 17
+dslr/images/mug/frame_0007.jpg 17
+dslr/images/mug/frame_0008.jpg 17
+dslr/images/tape_dispenser/frame_0001.jpg 29
+dslr/images/tape_dispenser/frame_0002.jpg 29
+dslr/images/tape_dispenser/frame_0003.jpg 29
+dslr/images/tape_dispenser/frame_0004.jpg 29
+dslr/images/tape_dispenser/frame_0005.jpg 29
+dslr/images/tape_dispenser/frame_0006.jpg 29
+dslr/images/tape_dispenser/frame_0007.jpg 29
+dslr/images/tape_dispenser/frame_0008.jpg 29
+dslr/images/tape_dispenser/frame_0009.jpg 29
+dslr/images/tape_dispenser/frame_0010.jpg 29
+dslr/images/tape_dispenser/frame_0011.jpg 29
+dslr/images/tape_dispenser/frame_0012.jpg 29
+dslr/images/tape_dispenser/frame_0013.jpg 29
+dslr/images/tape_dispenser/frame_0014.jpg 29
+dslr/images/tape_dispenser/frame_0015.jpg 29
+dslr/images/tape_dispenser/frame_0016.jpg 29
+dslr/images/tape_dispenser/frame_0017.jpg 29
+dslr/images/tape_dispenser/frame_0018.jpg 29
+dslr/images/tape_dispenser/frame_0019.jpg 29
+dslr/images/tape_dispenser/frame_0020.jpg 29
+dslr/images/tape_dispenser/frame_0021.jpg 29
+dslr/images/tape_dispenser/frame_0022.jpg 29
+dslr/images/pen/frame_0001.jpg 19
+dslr/images/pen/frame_0002.jpg 19
+dslr/images/pen/frame_0003.jpg 19
+dslr/images/pen/frame_0004.jpg 19
+dslr/images/pen/frame_0005.jpg 19
+dslr/images/pen/frame_0006.jpg 19
+dslr/images/pen/frame_0007.jpg 19
+dslr/images/pen/frame_0008.jpg 19
+dslr/images/pen/frame_0009.jpg 19
+dslr/images/pen/frame_0010.jpg 19
+dslr/images/bike/frame_0001.jpg 1
+dslr/images/bike/frame_0002.jpg 1
+dslr/images/bike/frame_0003.jpg 1
+dslr/images/bike/frame_0004.jpg 1
+dslr/images/bike/frame_0005.jpg 1
+dslr/images/bike/frame_0006.jpg 1
+dslr/images/bike/frame_0007.jpg 1
+dslr/images/bike/frame_0008.jpg 1
+dslr/images/bike/frame_0009.jpg 1
+dslr/images/bike/frame_0010.jpg 1
+dslr/images/bike/frame_0011.jpg 1
+dslr/images/bike/frame_0012.jpg 1
+dslr/images/bike/frame_0013.jpg 1
+dslr/images/bike/frame_0014.jpg 1
+dslr/images/bike/frame_0015.jpg 1
+dslr/images/bike/frame_0016.jpg 1
+dslr/images/bike/frame_0017.jpg 1
+dslr/images/bike/frame_0018.jpg 1
+dslr/images/bike/frame_0019.jpg 1
+dslr/images/bike/frame_0020.jpg 1
+dslr/images/bike/frame_0021.jpg 1
+dslr/images/punchers/frame_0001.jpg 23
+dslr/images/punchers/frame_0002.jpg 23
+dslr/images/punchers/frame_0003.jpg 23
+dslr/images/punchers/frame_0004.jpg 23
+dslr/images/punchers/frame_0005.jpg 23
+dslr/images/punchers/frame_0006.jpg 23
+dslr/images/punchers/frame_0007.jpg 23
+dslr/images/punchers/frame_0008.jpg 23
+dslr/images/punchers/frame_0009.jpg 23
+dslr/images/punchers/frame_0010.jpg 23
+dslr/images/punchers/frame_0011.jpg 23
+dslr/images/punchers/frame_0012.jpg 23
+dslr/images/punchers/frame_0013.jpg 23
+dslr/images/punchers/frame_0014.jpg 23
+dslr/images/punchers/frame_0015.jpg 23
+dslr/images/punchers/frame_0016.jpg 23
+dslr/images/punchers/frame_0017.jpg 23
+dslr/images/punchers/frame_0018.jpg 23
+dslr/images/back_pack/frame_0001.jpg 0
+dslr/images/back_pack/frame_0002.jpg 0
+dslr/images/back_pack/frame_0003.jpg 0
+dslr/images/back_pack/frame_0004.jpg 0
+dslr/images/back_pack/frame_0005.jpg 0
+dslr/images/back_pack/frame_0006.jpg 0
+dslr/images/back_pack/frame_0007.jpg 0
+dslr/images/back_pack/frame_0008.jpg 0
+dslr/images/back_pack/frame_0009.jpg 0
+dslr/images/back_pack/frame_0010.jpg 0
+dslr/images/back_pack/frame_0011.jpg 0
+dslr/images/back_pack/frame_0012.jpg 0
+dslr/images/desktop_computer/frame_0001.jpg 8
+dslr/images/desktop_computer/frame_0002.jpg 8
+dslr/images/desktop_computer/frame_0003.jpg 8
+dslr/images/desktop_computer/frame_0004.jpg 8
+dslr/images/desktop_computer/frame_0005.jpg 8
+dslr/images/desktop_computer/frame_0006.jpg 8
+dslr/images/desktop_computer/frame_0007.jpg 8
+dslr/images/desktop_computer/frame_0008.jpg 8
+dslr/images/desktop_computer/frame_0009.jpg 8
+dslr/images/desktop_computer/frame_0010.jpg 8
+dslr/images/desktop_computer/frame_0011.jpg 8
+dslr/images/desktop_computer/frame_0012.jpg 8
+dslr/images/desktop_computer/frame_0013.jpg 8
+dslr/images/desktop_computer/frame_0014.jpg 8
+dslr/images/desktop_computer/frame_0015.jpg 8
+dslr/images/speaker/frame_0001.jpg 27
+dslr/images/speaker/frame_0002.jpg 27
+dslr/images/speaker/frame_0003.jpg 27
+dslr/images/speaker/frame_0004.jpg 27
+dslr/images/speaker/frame_0005.jpg 27
+dslr/images/speaker/frame_0006.jpg 27
+dslr/images/speaker/frame_0007.jpg 27
+dslr/images/speaker/frame_0008.jpg 27
+dslr/images/speaker/frame_0009.jpg 27
+dslr/images/speaker/frame_0010.jpg 27
+dslr/images/speaker/frame_0011.jpg 27
+dslr/images/speaker/frame_0012.jpg 27
+dslr/images/speaker/frame_0013.jpg 27
+dslr/images/speaker/frame_0014.jpg 27
+dslr/images/speaker/frame_0015.jpg 27
+dslr/images/speaker/frame_0016.jpg 27
+dslr/images/speaker/frame_0017.jpg 27
+dslr/images/speaker/frame_0018.jpg 27
+dslr/images/speaker/frame_0019.jpg 27
+dslr/images/speaker/frame_0020.jpg 27
+dslr/images/speaker/frame_0021.jpg 27
+dslr/images/speaker/frame_0022.jpg 27
+dslr/images/speaker/frame_0023.jpg 27
+dslr/images/speaker/frame_0024.jpg 27
+dslr/images/speaker/frame_0025.jpg 27
+dslr/images/speaker/frame_0026.jpg 27
+dslr/images/mobile_phone/frame_0001.jpg 14
+dslr/images/mobile_phone/frame_0002.jpg 14
+dslr/images/mobile_phone/frame_0003.jpg 14
+dslr/images/mobile_phone/frame_0004.jpg 14
+dslr/images/mobile_phone/frame_0005.jpg 14
+dslr/images/mobile_phone/frame_0006.jpg 14
+dslr/images/mobile_phone/frame_0007.jpg 14
+dslr/images/mobile_phone/frame_0008.jpg 14
+dslr/images/mobile_phone/frame_0009.jpg 14
+dslr/images/mobile_phone/frame_0010.jpg 14
+dslr/images/mobile_phone/frame_0011.jpg 14
+dslr/images/mobile_phone/frame_0012.jpg 14
+dslr/images/mobile_phone/frame_0013.jpg 14
+dslr/images/mobile_phone/frame_0014.jpg 14
+dslr/images/mobile_phone/frame_0015.jpg 14
+dslr/images/mobile_phone/frame_0016.jpg 14
+dslr/images/mobile_phone/frame_0017.jpg 14
+dslr/images/mobile_phone/frame_0018.jpg 14
+dslr/images/mobile_phone/frame_0019.jpg 14
+dslr/images/mobile_phone/frame_0020.jpg 14
+dslr/images/mobile_phone/frame_0021.jpg 14
+dslr/images/mobile_phone/frame_0022.jpg 14
+dslr/images/mobile_phone/frame_0023.jpg 14
+dslr/images/mobile_phone/frame_0024.jpg 14
+dslr/images/mobile_phone/frame_0025.jpg 14
+dslr/images/mobile_phone/frame_0026.jpg 14
+dslr/images/mobile_phone/frame_0027.jpg 14
+dslr/images/mobile_phone/frame_0028.jpg 14
+dslr/images/mobile_phone/frame_0029.jpg 14
+dslr/images/mobile_phone/frame_0030.jpg 14
+dslr/images/mobile_phone/frame_0031.jpg 14
+dslr/images/paper_notebook/frame_0001.jpg 18
+dslr/images/paper_notebook/frame_0002.jpg 18
+dslr/images/paper_notebook/frame_0003.jpg 18
+dslr/images/paper_notebook/frame_0004.jpg 18
+dslr/images/paper_notebook/frame_0005.jpg 18
+dslr/images/paper_notebook/frame_0006.jpg 18
+dslr/images/paper_notebook/frame_0007.jpg 18
+dslr/images/paper_notebook/frame_0008.jpg 18
+dslr/images/paper_notebook/frame_0009.jpg 18
+dslr/images/paper_notebook/frame_0010.jpg 18
+dslr/images/ruler/frame_0001.jpg 25
+dslr/images/ruler/frame_0002.jpg 25
+dslr/images/ruler/frame_0003.jpg 25
+dslr/images/ruler/frame_0004.jpg 25
+dslr/images/ruler/frame_0005.jpg 25
+dslr/images/ruler/frame_0006.jpg 25
+dslr/images/ruler/frame_0007.jpg 25
+dslr/images/letter_tray/frame_0001.jpg 13
+dslr/images/letter_tray/frame_0002.jpg 13
+dslr/images/letter_tray/frame_0003.jpg 13
+dslr/images/letter_tray/frame_0004.jpg 13
+dslr/images/letter_tray/frame_0005.jpg 13
+dslr/images/letter_tray/frame_0006.jpg 13
+dslr/images/letter_tray/frame_0007.jpg 13
+dslr/images/letter_tray/frame_0008.jpg 13
+dslr/images/letter_tray/frame_0009.jpg 13
+dslr/images/letter_tray/frame_0010.jpg 13
+dslr/images/letter_tray/frame_0011.jpg 13
+dslr/images/letter_tray/frame_0012.jpg 13
+dslr/images/letter_tray/frame_0013.jpg 13
+dslr/images/letter_tray/frame_0014.jpg 13
+dslr/images/letter_tray/frame_0015.jpg 13
+dslr/images/letter_tray/frame_0016.jpg 13
+dslr/images/file_cabinet/frame_0001.jpg 9
+dslr/images/file_cabinet/frame_0002.jpg 9
+dslr/images/file_cabinet/frame_0003.jpg 9
+dslr/images/file_cabinet/frame_0004.jpg 9
+dslr/images/file_cabinet/frame_0005.jpg 9
+dslr/images/file_cabinet/frame_0006.jpg 9
+dslr/images/file_cabinet/frame_0007.jpg 9
+dslr/images/file_cabinet/frame_0008.jpg 9
+dslr/images/file_cabinet/frame_0009.jpg 9
+dslr/images/file_cabinet/frame_0010.jpg 9
+dslr/images/file_cabinet/frame_0011.jpg 9
+dslr/images/file_cabinet/frame_0012.jpg 9
+dslr/images/file_cabinet/frame_0013.jpg 9
+dslr/images/file_cabinet/frame_0014.jpg 9
+dslr/images/file_cabinet/frame_0015.jpg 9
+dslr/images/phone/frame_0001.jpg 20
+dslr/images/phone/frame_0002.jpg 20
+dslr/images/phone/frame_0003.jpg 20
+dslr/images/phone/frame_0004.jpg 20
+dslr/images/phone/frame_0005.jpg 20
+dslr/images/phone/frame_0006.jpg 20
+dslr/images/phone/frame_0007.jpg 20
+dslr/images/phone/frame_0008.jpg 20
+dslr/images/phone/frame_0009.jpg 20
+dslr/images/phone/frame_0010.jpg 20
+dslr/images/phone/frame_0011.jpg 20
+dslr/images/phone/frame_0012.jpg 20
+dslr/images/phone/frame_0013.jpg 20
+dslr/images/bookcase/frame_0001.jpg 3
+dslr/images/bookcase/frame_0002.jpg 3
+dslr/images/bookcase/frame_0003.jpg 3
+dslr/images/bookcase/frame_0004.jpg 3
+dslr/images/bookcase/frame_0005.jpg 3
+dslr/images/bookcase/frame_0006.jpg 3
+dslr/images/bookcase/frame_0007.jpg 3
+dslr/images/bookcase/frame_0008.jpg 3
+dslr/images/bookcase/frame_0009.jpg 3
+dslr/images/bookcase/frame_0010.jpg 3
+dslr/images/bookcase/frame_0011.jpg 3
+dslr/images/bookcase/frame_0012.jpg 3
+dslr/images/projector/frame_0001.jpg 22
+dslr/images/projector/frame_0002.jpg 22
+dslr/images/projector/frame_0003.jpg 22
+dslr/images/projector/frame_0004.jpg 22
+dslr/images/projector/frame_0005.jpg 22
+dslr/images/projector/frame_0006.jpg 22
+dslr/images/projector/frame_0007.jpg 22
+dslr/images/projector/frame_0008.jpg 22
+dslr/images/projector/frame_0009.jpg 22
+dslr/images/projector/frame_0010.jpg 22
+dslr/images/projector/frame_0011.jpg 22
+dslr/images/projector/frame_0012.jpg 22
+dslr/images/projector/frame_0013.jpg 22
+dslr/images/projector/frame_0014.jpg 22
+dslr/images/projector/frame_0015.jpg 22
+dslr/images/projector/frame_0016.jpg 22
+dslr/images/projector/frame_0017.jpg 22
+dslr/images/projector/frame_0018.jpg 22
+dslr/images/projector/frame_0019.jpg 22
+dslr/images/projector/frame_0020.jpg 22
+dslr/images/projector/frame_0021.jpg 22
+dslr/images/projector/frame_0022.jpg 22
+dslr/images/projector/frame_0023.jpg 22
+dslr/images/stapler/frame_0001.jpg 28
+dslr/images/stapler/frame_0002.jpg 28
+dslr/images/stapler/frame_0003.jpg 28
+dslr/images/stapler/frame_0004.jpg 28
+dslr/images/stapler/frame_0005.jpg 28
+dslr/images/stapler/frame_0006.jpg 28
+dslr/images/stapler/frame_0007.jpg 28
+dslr/images/stapler/frame_0008.jpg 28
+dslr/images/stapler/frame_0009.jpg 28
+dslr/images/stapler/frame_0010.jpg 28
+dslr/images/stapler/frame_0011.jpg 28
+dslr/images/stapler/frame_0012.jpg 28
+dslr/images/stapler/frame_0013.jpg 28
+dslr/images/stapler/frame_0014.jpg 28
+dslr/images/stapler/frame_0015.jpg 28
+dslr/images/stapler/frame_0016.jpg 28
+dslr/images/stapler/frame_0017.jpg 28
+dslr/images/stapler/frame_0018.jpg 28
+dslr/images/stapler/frame_0019.jpg 28
+dslr/images/stapler/frame_0020.jpg 28
+dslr/images/stapler/frame_0021.jpg 28
+dslr/images/trash_can/frame_0001.jpg 30
+dslr/images/trash_can/frame_0002.jpg 30
+dslr/images/trash_can/frame_0003.jpg 30
+dslr/images/trash_can/frame_0004.jpg 30
+dslr/images/trash_can/frame_0005.jpg 30
+dslr/images/trash_can/frame_0006.jpg 30
+dslr/images/trash_can/frame_0007.jpg 30
+dslr/images/trash_can/frame_0008.jpg 30
+dslr/images/trash_can/frame_0009.jpg 30
+dslr/images/trash_can/frame_0010.jpg 30
+dslr/images/trash_can/frame_0011.jpg 30
+dslr/images/trash_can/frame_0012.jpg 30
+dslr/images/trash_can/frame_0013.jpg 30
+dslr/images/trash_can/frame_0014.jpg 30
+dslr/images/trash_can/frame_0015.jpg 30
+dslr/images/bike_helmet/frame_0001.jpg 2
+dslr/images/bike_helmet/frame_0002.jpg 2
+dslr/images/bike_helmet/frame_0003.jpg 2
+dslr/images/bike_helmet/frame_0004.jpg 2
+dslr/images/bike_helmet/frame_0005.jpg 2
+dslr/images/bike_helmet/frame_0006.jpg 2
+dslr/images/bike_helmet/frame_0007.jpg 2
+dslr/images/bike_helmet/frame_0008.jpg 2
+dslr/images/bike_helmet/frame_0009.jpg 2
+dslr/images/bike_helmet/frame_0010.jpg 2
+dslr/images/bike_helmet/frame_0011.jpg 2
+dslr/images/bike_helmet/frame_0012.jpg 2
+dslr/images/bike_helmet/frame_0013.jpg 2
+dslr/images/bike_helmet/frame_0014.jpg 2
+dslr/images/bike_helmet/frame_0015.jpg 2
+dslr/images/bike_helmet/frame_0016.jpg 2
+dslr/images/bike_helmet/frame_0017.jpg 2
+dslr/images/bike_helmet/frame_0018.jpg 2
+dslr/images/bike_helmet/frame_0019.jpg 2
+dslr/images/bike_helmet/frame_0020.jpg 2
+dslr/images/bike_helmet/frame_0021.jpg 2
+dslr/images/bike_helmet/frame_0022.jpg 2
+dslr/images/bike_helmet/frame_0023.jpg 2
+dslr/images/bike_helmet/frame_0024.jpg 2
+dslr/images/headphones/frame_0001.jpg 10
+dslr/images/headphones/frame_0002.jpg 10
+dslr/images/headphones/frame_0003.jpg 10
+dslr/images/headphones/frame_0004.jpg 10
+dslr/images/headphones/frame_0005.jpg 10
+dslr/images/headphones/frame_0006.jpg 10
+dslr/images/headphones/frame_0007.jpg 10
+dslr/images/headphones/frame_0008.jpg 10
+dslr/images/headphones/frame_0009.jpg 10
+dslr/images/headphones/frame_0010.jpg 10
+dslr/images/headphones/frame_0011.jpg 10
+dslr/images/headphones/frame_0012.jpg 10
+dslr/images/headphones/frame_0013.jpg 10
+dslr/images/desk_lamp/frame_0001.jpg 7
+dslr/images/desk_lamp/frame_0002.jpg 7
+dslr/images/desk_lamp/frame_0003.jpg 7
+dslr/images/desk_lamp/frame_0004.jpg 7
+dslr/images/desk_lamp/frame_0005.jpg 7
+dslr/images/desk_lamp/frame_0006.jpg 7
+dslr/images/desk_lamp/frame_0007.jpg 7
+dslr/images/desk_lamp/frame_0008.jpg 7
+dslr/images/desk_lamp/frame_0009.jpg 7
+dslr/images/desk_lamp/frame_0010.jpg 7
+dslr/images/desk_lamp/frame_0011.jpg 7
+dslr/images/desk_lamp/frame_0012.jpg 7
+dslr/images/desk_lamp/frame_0013.jpg 7
+dslr/images/desk_lamp/frame_0014.jpg 7
+dslr/images/desk_chair/frame_0001.jpg 6
+dslr/images/desk_chair/frame_0002.jpg 6
+dslr/images/desk_chair/frame_0003.jpg 6
+dslr/images/desk_chair/frame_0004.jpg 6
+dslr/images/desk_chair/frame_0005.jpg 6
+dslr/images/desk_chair/frame_0006.jpg 6
+dslr/images/desk_chair/frame_0007.jpg 6
+dslr/images/desk_chair/frame_0008.jpg 6
+dslr/images/desk_chair/frame_0009.jpg 6
+dslr/images/desk_chair/frame_0010.jpg 6
+dslr/images/desk_chair/frame_0011.jpg 6
+dslr/images/desk_chair/frame_0012.jpg 6
+dslr/images/desk_chair/frame_0013.jpg 6
+dslr/images/bottle/frame_0001.jpg 4
+dslr/images/bottle/frame_0002.jpg 4
+dslr/images/bottle/frame_0003.jpg 4
+dslr/images/bottle/frame_0004.jpg 4
+dslr/images/bottle/frame_0005.jpg 4
+dslr/images/bottle/frame_0006.jpg 4
+dslr/images/bottle/frame_0007.jpg 4
+dslr/images/bottle/frame_0008.jpg 4
+dslr/images/bottle/frame_0009.jpg 4
+dslr/images/bottle/frame_0010.jpg 4
+dslr/images/bottle/frame_0011.jpg 4
+dslr/images/bottle/frame_0012.jpg 4
+dslr/images/bottle/frame_0013.jpg 4
+dslr/images/bottle/frame_0014.jpg 4
+dslr/images/bottle/frame_0015.jpg 4
+dslr/images/bottle/frame_0016.jpg 4
--- a/data/office-31/webcam_list.txt
+++ b/data/office-31/webcam_list.txt
@ -0,0 +1,795 @@
+webcam/images/calculator/frame_0001.jpg 5
+webcam/images/calculator/frame_0002.jpg 5
+webcam/images/calculator/frame_0003.jpg 5
+webcam/images/calculator/frame_0004.jpg 5
+webcam/images/calculator/frame_0005.jpg 5
+webcam/images/calculator/frame_0006.jpg 5
+webcam/images/calculator/frame_0007.jpg 5
+webcam/images/calculator/frame_0008.jpg 5
+webcam/images/calculator/frame_0009.jpg 5
+webcam/images/calculator/frame_0010.jpg 5
+webcam/images/calculator/frame_0011.jpg 5
+webcam/images/calculator/frame_0012.jpg 5
+webcam/images/calculator/frame_0013.jpg 5
+webcam/images/calculator/frame_0014.jpg 5
+webcam/images/calculator/frame_0015.jpg 5
+webcam/images/calculator/frame_0016.jpg 5
+webcam/images/calculator/frame_0017.jpg 5
+webcam/images/calculator/frame_0018.jpg 5
+webcam/images/calculator/frame_0019.jpg 5
+webcam/images/calculator/frame_0020.jpg 5
+webcam/images/calculator/frame_0021.jpg 5
+webcam/images/calculator/frame_0022.jpg 5
+webcam/images/calculator/frame_0023.jpg 5
+webcam/images/calculator/frame_0024.jpg 5
+webcam/images/calculator/frame_0025.jpg 5
+webcam/images/calculator/frame_0026.jpg 5
+webcam/images/calculator/frame_0027.jpg 5
+webcam/images/calculator/frame_0028.jpg 5
+webcam/images/calculator/frame_0029.jpg 5
+webcam/images/calculator/frame_0030.jpg 5
+webcam/images/calculator/frame_0031.jpg 5
+webcam/images/ring_binder/frame_0001.jpg 24
+webcam/images/ring_binder/frame_0002.jpg 24
+webcam/images/ring_binder/frame_0003.jpg 24
+webcam/images/ring_binder/frame_0004.jpg 24
+webcam/images/ring_binder/frame_0005.jpg 24
+webcam/images/ring_binder/frame_0006.jpg 24
+webcam/images/ring_binder/frame_0007.jpg 24
+webcam/images/ring_binder/frame_0008.jpg 24
+webcam/images/ring_binder/frame_0009.jpg 24
+webcam/images/ring_binder/frame_0010.jpg 24
+webcam/images/ring_binder/frame_0011.jpg 24
+webcam/images/ring_binder/frame_0012.jpg 24
+webcam/images/ring_binder/frame_0013.jpg 24
+webcam/images/ring_binder/frame_0014.jpg 24
+webcam/images/ring_binder/frame_0015.jpg 24
+webcam/images/ring_binder/frame_0016.jpg 24
+webcam/images/ring_binder/frame_0017.jpg 24
+webcam/images/ring_binder/frame_0018.jpg 24
+webcam/images/ring_binder/frame_0019.jpg 24
+webcam/images/ring_binder/frame_0020.jpg 24
+webcam/images/ring_binder/frame_0021.jpg 24
+webcam/images/ring_binder/frame_0022.jpg 24
+webcam/images/ring_binder/frame_0023.jpg 24
+webcam/images/ring_binder/frame_0024.jpg 24
+webcam/images/ring_binder/frame_0025.jpg 24
+webcam/images/ring_binder/frame_0026.jpg 24
+webcam/images/ring_binder/frame_0027.jpg 24
+webcam/images/ring_binder/frame_0028.jpg 24
+webcam/images/ring_binder/frame_0029.jpg 24
+webcam/images/ring_binder/frame_0030.jpg 24
+webcam/images/ring_binder/frame_0031.jpg 24
+webcam/images/ring_binder/frame_0032.jpg 24
+webcam/images/ring_binder/frame_0033.jpg 24
+webcam/images/ring_binder/frame_0034.jpg 24
+webcam/images/ring_binder/frame_0035.jpg 24
+webcam/images/ring_binder/frame_0036.jpg 24
+webcam/images/ring_binder/frame_0037.jpg 24
+webcam/images/ring_binder/frame_0038.jpg 24
+webcam/images/ring_binder/frame_0039.jpg 24
+webcam/images/ring_binder/frame_0040.jpg 24
+webcam/images/printer/frame_0001.jpg 21
+webcam/images/printer/frame_0002.jpg 21
+webcam/images/printer/frame_0003.jpg 21
+webcam/images/printer/frame_0004.jpg 21
+webcam/images/printer/frame_0005.jpg 21
+webcam/images/printer/frame_0006.jpg 21
+webcam/images/printer/frame_0007.jpg 21
+webcam/images/printer/frame_0008.jpg 21
+webcam/images/printer/frame_0009.jpg 21
+webcam/images/printer/frame_0010.jpg 21
+webcam/images/printer/frame_0011.jpg 21
+webcam/images/printer/frame_0012.jpg 21
+webcam/images/printer/frame_0013.jpg 21
+webcam/images/printer/frame_0014.jpg 21
+webcam/images/printer/frame_0015.jpg 21
+webcam/images/printer/frame_0016.jpg 21
+webcam/images/printer/frame_0017.jpg 21
+webcam/images/printer/frame_0018.jpg 21
+webcam/images/printer/frame_0019.jpg 21
+webcam/images/printer/frame_0020.jpg 21
+webcam/images/keyboard/frame_0001.jpg 11
+webcam/images/keyboard/frame_0002.jpg 11
+webcam/images/keyboard/frame_0003.jpg 11
+webcam/images/keyboard/frame_0004.jpg 11
+webcam/images/keyboard/frame_0005.jpg 11
+webcam/images/keyboard/frame_0006.jpg 11
+webcam/images/keyboard/frame_0007.jpg 11
+webcam/images/keyboard/frame_0008.jpg 11
+webcam/images/keyboard/frame_0009.jpg 11
+webcam/images/keyboard/frame_0010.jpg 11
+webcam/images/keyboard/frame_0011.jpg 11
+webcam/images/keyboard/frame_0012.jpg 11
+webcam/images/keyboard/frame_0013.jpg 11
+webcam/images/keyboard/frame_0014.jpg 11
+webcam/images/keyboard/frame_0015.jpg 11
+webcam/images/keyboard/frame_0016.jpg 11
+webcam/images/keyboard/frame_0017.jpg 11
+webcam/images/keyboard/frame_0018.jpg 11
+webcam/images/keyboard/frame_0019.jpg 11
+webcam/images/keyboard/frame_0020.jpg 11
+webcam/images/keyboard/frame_0021.jpg 11
+webcam/images/keyboard/frame_0022.jpg 11
+webcam/images/keyboard/frame_0023.jpg 11
+webcam/images/keyboard/frame_0024.jpg 11
+webcam/images/keyboard/frame_0025.jpg 11
+webcam/images/keyboard/frame_0026.jpg 11
+webcam/images/keyboard/frame_0027.jpg 11
+webcam/images/scissors/frame_0001.jpg 26
+webcam/images/scissors/frame_0002.jpg 26
+webcam/images/scissors/frame_0003.jpg 26
+webcam/images/scissors/frame_0004.jpg 26
+webcam/images/scissors/frame_0005.jpg 26
+webcam/images/scissors/frame_0006.jpg 26
+webcam/images/scissors/frame_0007.jpg 26
+webcam/images/scissors/frame_0008.jpg 26
+webcam/images/scissors/frame_0009.jpg 26
+webcam/images/scissors/frame_0010.jpg 26
+webcam/images/scissors/frame_0011.jpg 26
+webcam/images/scissors/frame_0012.jpg 26
+webcam/images/scissors/frame_0013.jpg 26
+webcam/images/scissors/frame_0014.jpg 26
+webcam/images/scissors/frame_0015.jpg 26
+webcam/images/scissors/frame_0016.jpg 26
+webcam/images/scissors/frame_0017.jpg 26
+webcam/images/scissors/frame_0018.jpg 26
+webcam/images/scissors/frame_0019.jpg 26
+webcam/images/scissors/frame_0020.jpg 26
+webcam/images/scissors/frame_0021.jpg 26
+webcam/images/scissors/frame_0022.jpg 26
+webcam/images/scissors/frame_0023.jpg 26
+webcam/images/scissors/frame_0024.jpg 26
+webcam/images/scissors/frame_0025.jpg 26
+webcam/images/laptop_computer/frame_0001.jpg 12
+webcam/images/laptop_computer/frame_0002.jpg 12
+webcam/images/laptop_computer/frame_0003.jpg 12
+webcam/images/laptop_computer/frame_0004.jpg 12
+webcam/images/laptop_computer/frame_0005.jpg 12
+webcam/images/laptop_computer/frame_0006.jpg 12
+webcam/images/laptop_computer/frame_0007.jpg 12
+webcam/images/laptop_computer/frame_0008.jpg 12
+webcam/images/laptop_computer/frame_0009.jpg 12
+webcam/images/laptop_computer/frame_0010.jpg 12
+webcam/images/laptop_computer/frame_0011.jpg 12
+webcam/images/laptop_computer/frame_0012.jpg 12
+webcam/images/laptop_computer/frame_0013.jpg 12
+webcam/images/laptop_computer/frame_0014.jpg 12
+webcam/images/laptop_computer/frame_0015.jpg 12
+webcam/images/laptop_computer/frame_0016.jpg 12
+webcam/images/laptop_computer/frame_0017.jpg 12
+webcam/images/laptop_computer/frame_0018.jpg 12
+webcam/images/laptop_computer/frame_0019.jpg 12
+webcam/images/laptop_computer/frame_0020.jpg 12
+webcam/images/laptop_computer/frame_0021.jpg 12
+webcam/images/laptop_computer/frame_0022.jpg 12
+webcam/images/laptop_computer/frame_0023.jpg 12
+webcam/images/laptop_computer/frame_0024.jpg 12
+webcam/images/laptop_computer/frame_0025.jpg 12
+webcam/images/laptop_computer/frame_0026.jpg 12
+webcam/images/laptop_computer/frame_0027.jpg 12
+webcam/images/laptop_computer/frame_0028.jpg 12
+webcam/images/laptop_computer/frame_0029.jpg 12
+webcam/images/laptop_computer/frame_0030.jpg 12
+webcam/images/mouse/frame_0001.jpg 16
+webcam/images/mouse/frame_0002.jpg 16
+webcam/images/mouse/frame_0003.jpg 16
+webcam/images/mouse/frame_0004.jpg 16
+webcam/images/mouse/frame_0005.jpg 16
+webcam/images/mouse/frame_0006.jpg 16
+webcam/images/mouse/frame_0007.jpg 16
+webcam/images/mouse/frame_0008.jpg 16
+webcam/images/mouse/frame_0009.jpg 16
+webcam/images/mouse/frame_0010.jpg 16
+webcam/images/mouse/frame_0011.jpg 16
+webcam/images/mouse/frame_0012.jpg 16
+webcam/images/mouse/frame_0013.jpg 16
+webcam/images/mouse/frame_0014.jpg 16
+webcam/images/mouse/frame_0015.jpg 16
+webcam/images/mouse/frame_0016.jpg 16
+webcam/images/mouse/frame_0017.jpg 16
+webcam/images/mouse/frame_0018.jpg 16
+webcam/images/mouse/frame_0019.jpg 16
+webcam/images/mouse/frame_0020.jpg 16
+webcam/images/mouse/frame_0021.jpg 16
+webcam/images/mouse/frame_0022.jpg 16
+webcam/images/mouse/frame_0023.jpg 16
+webcam/images/mouse/frame_0024.jpg 16
+webcam/images/mouse/frame_0025.jpg 16
+webcam/images/mouse/frame_0026.jpg 16
+webcam/images/mouse/frame_0027.jpg 16
+webcam/images/mouse/frame_0028.jpg 16
+webcam/images/mouse/frame_0029.jpg 16
+webcam/images/mouse/frame_0030.jpg 16
+webcam/images/monitor/frame_0001.jpg 15
+webcam/images/monitor/frame_0002.jpg 15
+webcam/images/monitor/frame_0003.jpg 15
+webcam/images/monitor/frame_0004.jpg 15
+webcam/images/monitor/frame_0005.jpg 15
+webcam/images/monitor/frame_0006.jpg 15
+webcam/images/monitor/frame_0007.jpg 15
+webcam/images/monitor/frame_0008.jpg 15
+webcam/images/monitor/frame_0009.jpg 15
+webcam/images/monitor/frame_0010.jpg 15
+webcam/images/monitor/frame_0011.jpg 15
+webcam/images/monitor/frame_0012.jpg 15
+webcam/images/monitor/frame_0013.jpg 15
+webcam/images/monitor/frame_0014.jpg 15
+webcam/images/monitor/frame_0015.jpg 15
+webcam/images/monitor/frame_0016.jpg 15
+webcam/images/monitor/frame_0017.jpg 15
+webcam/images/monitor/frame_0018.jpg 15
+webcam/images/monitor/frame_0019.jpg 15
+webcam/images/monitor/frame_0020.jpg 15
+webcam/images/monitor/frame_0021.jpg 15
+webcam/images/monitor/frame_0022.jpg 15
+webcam/images/monitor/frame_0023.jpg 15
+webcam/images/monitor/frame_0024.jpg 15
+webcam/images/monitor/frame_0025.jpg 15
+webcam/images/monitor/frame_0026.jpg 15
+webcam/images/monitor/frame_0027.jpg 15
+webcam/images/monitor/frame_0028.jpg 15
+webcam/images/monitor/frame_0029.jpg 15
+webcam/images/monitor/frame_0030.jpg 15
+webcam/images/monitor/frame_0031.jpg 15
+webcam/images/monitor/frame_0032.jpg 15
+webcam/images/monitor/frame_0033.jpg 15
+webcam/images/monitor/frame_0034.jpg 15
+webcam/images/monitor/frame_0035.jpg 15
+webcam/images/monitor/frame_0036.jpg 15
+webcam/images/monitor/frame_0037.jpg 15
+webcam/images/monitor/frame_0038.jpg 15
+webcam/images/monitor/frame_0039.jpg 15
+webcam/images/monitor/frame_0040.jpg 15
+webcam/images/monitor/frame_0041.jpg 15
+webcam/images/monitor/frame_0042.jpg 15
+webcam/images/monitor/frame_0043.jpg 15
+webcam/images/mug/frame_0001.jpg 17
+webcam/images/mug/frame_0002.jpg 17
+webcam/images/mug/frame_0003.jpg 17
+webcam/images/mug/frame_0004.jpg 17
+webcam/images/mug/frame_0005.jpg 17
+webcam/images/mug/frame_0006.jpg 17
+webcam/images/mug/frame_0007.jpg 17
+webcam/images/mug/frame_0008.jpg 17
+webcam/images/mug/frame_0009.jpg 17
+webcam/images/mug/frame_0010.jpg 17
+webcam/images/mug/frame_0011.jpg 17
+webcam/images/mug/frame_0012.jpg 17
+webcam/images/mug/frame_0013.jpg 17
+webcam/images/mug/frame_0014.jpg 17
+webcam/images/mug/frame_0015.jpg 17
+webcam/images/mug/frame_0016.jpg 17
+webcam/images/mug/frame_0017.jpg 17
+webcam/images/mug/frame_0018.jpg 17
+webcam/images/mug/frame_0019.jpg 17
+webcam/images/mug/frame_0020.jpg 17
+webcam/images/mug/frame_0021.jpg 17
+webcam/images/mug/frame_0022.jpg 17
+webcam/images/mug/frame_0023.jpg 17
+webcam/images/mug/frame_0024.jpg 17
+webcam/images/mug/frame_0025.jpg 17
+webcam/images/mug/frame_0026.jpg 17
+webcam/images/mug/frame_0027.jpg 17
+webcam/images/tape_dispenser/frame_0001.jpg 29
+webcam/images/tape_dispenser/frame_0002.jpg 29
+webcam/images/tape_dispenser/frame_0003.jpg 29
+webcam/images/tape_dispenser/frame_0004.jpg 29
+webcam/images/tape_dispenser/frame_0005.jpg 29
+webcam/images/tape_dispenser/frame_0006.jpg 29
+webcam/images/tape_dispenser/frame_0007.jpg 29
+webcam/images/tape_dispenser/frame_0008.jpg 29
+webcam/images/tape_dispenser/frame_0009.jpg 29
+webcam/images/tape_dispenser/frame_0010.jpg 29
+webcam/images/tape_dispenser/frame_0011.jpg 29
+webcam/images/tape_dispenser/frame_0012.jpg 29
+webcam/images/tape_dispenser/frame_0013.jpg 29
+webcam/images/tape_dispenser/frame_0014.jpg 29
+webcam/images/tape_dispenser/frame_0015.jpg 29
+webcam/images/tape_dispenser/frame_0016.jpg 29
+webcam/images/tape_dispenser/frame_0017.jpg 29
+webcam/images/tape_dispenser/frame_0018.jpg 29
+webcam/images/tape_dispenser/frame_0019.jpg 29
+webcam/images/tape_dispenser/frame_0020.jpg 29
+webcam/images/tape_dispenser/frame_0021.jpg 29
+webcam/images/tape_dispenser/frame_0022.jpg 29
+webcam/images/tape_dispenser/frame_0023.jpg 29
+webcam/images/pen/frame_0001.jpg 19
+webcam/images/pen/frame_0002.jpg 19
+webcam/images/pen/frame_0003.jpg 19
+webcam/images/pen/frame_0004.jpg 19
+webcam/images/pen/frame_0005.jpg 19
+webcam/images/pen/frame_0006.jpg 19
+webcam/images/pen/frame_0007.jpg 19
+webcam/images/pen/frame_0008.jpg 19
+webcam/images/pen/frame_0009.jpg 19
+webcam/images/pen/frame_0010.jpg 19
+webcam/images/pen/frame_0011.jpg 19
+webcam/images/pen/frame_0012.jpg 19
+webcam/images/pen/frame_0013.jpg 19
+webcam/images/pen/frame_0014.jpg 19
+webcam/images/pen/frame_0015.jpg 19
+webcam/images/pen/frame_0016.jpg 19
+webcam/images/pen/frame_0017.jpg 19
+webcam/images/pen/frame_0018.jpg 19
+webcam/images/pen/frame_0019.jpg 19
+webcam/images/pen/frame_0020.jpg 19
+webcam/images/pen/frame_0021.jpg 19
+webcam/images/pen/frame_0022.jpg 19
+webcam/images/pen/frame_0023.jpg 19
+webcam/images/pen/frame_0024.jpg 19
+webcam/images/pen/frame_0025.jpg 19
+webcam/images/pen/frame_0026.jpg 19
+webcam/images/pen/frame_0027.jpg 19
+webcam/images/pen/frame_0028.jpg 19
+webcam/images/pen/frame_0029.jpg 19
+webcam/images/pen/frame_0030.jpg 19
+webcam/images/pen/frame_0031.jpg 19
+webcam/images/pen/frame_0032.jpg 19
+webcam/images/bike/frame_0001.jpg 1
+webcam/images/bike/frame_0002.jpg 1
+webcam/images/bike/frame_0003.jpg 1
+webcam/images/bike/frame_0004.jpg 1
+webcam/images/bike/frame_0005.jpg 1
+webcam/images/bike/frame_0006.jpg 1
+webcam/images/bike/frame_0007.jpg 1
+webcam/images/bike/frame_0008.jpg 1
+webcam/images/bike/frame_0009.jpg 1
+webcam/images/bike/frame_0010.jpg 1
+webcam/images/bike/frame_0011.jpg 1
+webcam/images/bike/frame_0012.jpg 1
+webcam/images/bike/frame_0013.jpg 1
+webcam/images/bike/frame_0014.jpg 1
+webcam/images/bike/frame_0015.jpg 1
+webcam/images/bike/frame_0016.jpg 1
+webcam/images/bike/frame_0017.jpg 1
+webcam/images/bike/frame_0018.jpg 1
+webcam/images/bike/frame_0019.jpg 1
+webcam/images/bike/frame_0020.jpg 1
+webcam/images/bike/frame_0021.jpg 1
+webcam/images/punchers/frame_0001.jpg 23
+webcam/images/punchers/frame_0002.jpg 23
+webcam/images/punchers/frame_0003.jpg 23
+webcam/images/punchers/frame_0004.jpg 23
+webcam/images/punchers/frame_0005.jpg 23
+webcam/images/punchers/frame_0006.jpg 23
+webcam/images/punchers/frame_0007.jpg 23
+webcam/images/punchers/frame_0008.jpg 23
+webcam/images/punchers/frame_0009.jpg 23
+webcam/images/punchers/frame_0010.jpg 23
+webcam/images/punchers/frame_0011.jpg 23
+webcam/images/punchers/frame_0012.jpg 23
+webcam/images/punchers/frame_0013.jpg 23
+webcam/images/punchers/frame_0014.jpg 23
+webcam/images/punchers/frame_0015.jpg 23
+webcam/images/punchers/frame_0016.jpg 23
+webcam/images/punchers/frame_0017.jpg 23
+webcam/images/punchers/frame_0018.jpg 23
+webcam/images/punchers/frame_0019.jpg 23
+webcam/images/punchers/frame_0020.jpg 23
+webcam/images/punchers/frame_0021.jpg 23
+webcam/images/punchers/frame_0022.jpg 23
+webcam/images/punchers/frame_0023.jpg 23
+webcam/images/punchers/frame_0024.jpg 23
+webcam/images/punchers/frame_0025.jpg 23
+webcam/images/punchers/frame_0026.jpg 23
+webcam/images/punchers/frame_0027.jpg 23
+webcam/images/back_pack/frame_0001.jpg 0
+webcam/images/back_pack/frame_0002.jpg 0
+webcam/images/back_pack/frame_0003.jpg 0
+webcam/images/back_pack/frame_0004.jpg 0
+webcam/images/back_pack/frame_0005.jpg 0
+webcam/images/back_pack/frame_0006.jpg 0
+webcam/images/back_pack/frame_0007.jpg 0
+webcam/images/back_pack/frame_0008.jpg 0
+webcam/images/back_pack/frame_0009.jpg 0
+webcam/images/back_pack/frame_0010.jpg 0
+webcam/images/back_pack/frame_0011.jpg 0
+webcam/images/back_pack/frame_0012.jpg 0
+webcam/images/back_pack/frame_0013.jpg 0
+webcam/images/back_pack/frame_0014.jpg 0
+webcam/images/back_pack/frame_0015.jpg 0
+webcam/images/back_pack/frame_0016.jpg 0
+webcam/images/back_pack/frame_0017.jpg 0
+webcam/images/back_pack/frame_0018.jpg 0
+webcam/images/back_pack/frame_0019.jpg 0
+webcam/images/back_pack/frame_0020.jpg 0
+webcam/images/back_pack/frame_0021.jpg 0
+webcam/images/back_pack/frame_0022.jpg 0
+webcam/images/back_pack/frame_0023.jpg 0
+webcam/images/back_pack/frame_0024.jpg 0
+webcam/images/back_pack/frame_0025.jpg 0
+webcam/images/back_pack/frame_0026.jpg 0
+webcam/images/back_pack/frame_0027.jpg 0
+webcam/images/back_pack/frame_0028.jpg 0
+webcam/images/back_pack/frame_0029.jpg 0
+webcam/images/desktop_computer/frame_0001.jpg 8
+webcam/images/desktop_computer/frame_0002.jpg 8
+webcam/images/desktop_computer/frame_0003.jpg 8
+webcam/images/desktop_computer/frame_0004.jpg 8
+webcam/images/desktop_computer/frame_0005.jpg 8
+webcam/images/desktop_computer/frame_0006.jpg 8
+webcam/images/desktop_computer/frame_0007.jpg 8
+webcam/images/desktop_computer/frame_0008.jpg 8
+webcam/images/desktop_computer/frame_0009.jpg 8
+webcam/images/desktop_computer/frame_0010.jpg 8
+webcam/images/desktop_computer/frame_0011.jpg 8
+webcam/images/desktop_computer/frame_0012.jpg 8
+webcam/images/desktop_computer/frame_0013.jpg 8
+webcam/images/desktop_computer/frame_0014.jpg 8
+webcam/images/desktop_computer/frame_0015.jpg 8
+webcam/images/desktop_computer/frame_0016.jpg 8
+webcam/images/desktop_computer/frame_0017.jpg 8
+webcam/images/desktop_computer/frame_0018.jpg 8
+webcam/images/desktop_computer/frame_0019.jpg 8
+webcam/images/desktop_computer/frame_0020.jpg 8
+webcam/images/desktop_computer/frame_0021.jpg 8
+webcam/images/speaker/frame_0001.jpg 27
+webcam/images/speaker/frame_0002.jpg 27
+webcam/images/speaker/frame_0003.jpg 27
+webcam/images/speaker/frame_0004.jpg 27
+webcam/images/speaker/frame_0005.jpg 27
+webcam/images/speaker/frame_0006.jpg 27
+webcam/images/speaker/frame_0007.jpg 27
+webcam/images/speaker/frame_0008.jpg 27
+webcam/images/speaker/frame_0009.jpg 27
+webcam/images/speaker/frame_0010.jpg 27
+webcam/images/speaker/frame_0011.jpg 27
+webcam/images/speaker/frame_0012.jpg 27
+webcam/images/speaker/frame_0013.jpg 27
+webcam/images/speaker/frame_0014.jpg 27
+webcam/images/speaker/frame_0015.jpg 27
+webcam/images/speaker/frame_0016.jpg 27
+webcam/images/speaker/frame_0017.jpg 27
+webcam/images/speaker/frame_0018.jpg 27
+webcam/images/speaker/frame_0019.jpg 27
+webcam/images/speaker/frame_0020.jpg 27
+webcam/images/speaker/frame_0021.jpg 27
+webcam/images/speaker/frame_0022.jpg 27
+webcam/images/speaker/frame_0023.jpg 27
+webcam/images/speaker/frame_0024.jpg 27
+webcam/images/speaker/frame_0025.jpg 27
+webcam/images/speaker/frame_0026.jpg 27
+webcam/images/speaker/frame_0027.jpg 27
+webcam/images/speaker/frame_0028.jpg 27
+webcam/images/speaker/frame_0029.jpg 27
+webcam/images/speaker/frame_0030.jpg 27
+webcam/images/mobile_phone/frame_0001.jpg 14
+webcam/images/mobile_phone/frame_0002.jpg 14
+webcam/images/mobile_phone/frame_0003.jpg 14
+webcam/images/mobile_phone/frame_0004.jpg 14
+webcam/images/mobile_phone/frame_0005.jpg 14
+webcam/images/mobile_phone/frame_0006.jpg 14
+webcam/images/mobile_phone/frame_0007.jpg 14
+webcam/images/mobile_phone/frame_0008.jpg 14
+webcam/images/mobile_phone/frame_0009.jpg 14
+webcam/images/mobile_phone/frame_0010.jpg 14
+webcam/images/mobile_phone/frame_0011.jpg 14
+webcam/images/mobile_phone/frame_0012.jpg 14
+webcam/images/mobile_phone/frame_0013.jpg 14
+webcam/images/mobile_phone/frame_0014.jpg 14
+webcam/images/mobile_phone/frame_0015.jpg 14
+webcam/images/mobile_phone/frame_0016.jpg 14
+webcam/images/mobile_phone/frame_0017.jpg 14
+webcam/images/mobile_phone/frame_0018.jpg 14
+webcam/images/mobile_phone/frame_0019.jpg 14
+webcam/images/mobile_phone/frame_0020.jpg 14
+webcam/images/mobile_phone/frame_0021.jpg 14
+webcam/images/mobile_phone/frame_0022.jpg 14
+webcam/images/mobile_phone/frame_0023.jpg 14
+webcam/images/mobile_phone/frame_0024.jpg 14
+webcam/images/mobile_phone/frame_0025.jpg 14
+webcam/images/mobile_phone/frame_0026.jpg 14
+webcam/images/mobile_phone/frame_0027.jpg 14
+webcam/images/mobile_phone/frame_0028.jpg 14
+webcam/images/mobile_phone/frame_0029.jpg 14
+webcam/images/mobile_phone/frame_0030.jpg 14
+webcam/images/paper_notebook/frame_0001.jpg 18
+webcam/images/paper_notebook/frame_0002.jpg 18
+webcam/images/paper_notebook/frame_0003.jpg 18
+webcam/images/paper_notebook/frame_0004.jpg 18
+webcam/images/paper_notebook/frame_0005.jpg 18
+webcam/images/paper_notebook/frame_0006.jpg 18
+webcam/images/paper_notebook/frame_0007.jpg 18
+webcam/images/paper_notebook/frame_0008.jpg 18
+webcam/images/paper_notebook/frame_0009.jpg 18
+webcam/images/paper_notebook/frame_0010.jpg 18
+webcam/images/paper_notebook/frame_0011.jpg 18
+webcam/images/paper_notebook/frame_0012.jpg 18
+webcam/images/paper_notebook/frame_0013.jpg 18
+webcam/images/paper_notebook/frame_0014.jpg 18
+webcam/images/paper_notebook/frame_0015.jpg 18
+webcam/images/paper_notebook/frame_0016.jpg 18
+webcam/images/paper_notebook/frame_0017.jpg 18
+webcam/images/paper_notebook/frame_0018.jpg 18
+webcam/images/paper_notebook/frame_0019.jpg 18
+webcam/images/paper_notebook/frame_0020.jpg 18
+webcam/images/paper_notebook/frame_0021.jpg 18
+webcam/images/paper_notebook/frame_0022.jpg 18
+webcam/images/paper_notebook/frame_0023.jpg 18
+webcam/images/paper_notebook/frame_0024.jpg 18
+webcam/images/paper_notebook/frame_0025.jpg 18
+webcam/images/paper_notebook/frame_0026.jpg 18
+webcam/images/paper_notebook/frame_0027.jpg 18
+webcam/images/paper_notebook/frame_0028.jpg 18
+webcam/images/ruler/frame_0001.jpg 25
+webcam/images/ruler/frame_0002.jpg 25
+webcam/images/ruler/frame_0003.jpg 25
+webcam/images/ruler/frame_0004.jpg 25
+webcam/images/ruler/frame_0005.jpg 25
+webcam/images/ruler/frame_0006.jpg 25
+webcam/images/ruler/frame_0007.jpg 25
+webcam/images/ruler/frame_0008.jpg 25
+webcam/images/ruler/frame_0009.jpg 25
+webcam/images/ruler/frame_0010.jpg 25
+webcam/images/ruler/frame_0011.jpg 25
+webcam/images/letter_tray/frame_0001.jpg 13
+webcam/images/letter_tray/frame_0002.jpg 13
+webcam/images/letter_tray/frame_0003.jpg 13
+webcam/images/letter_tray/frame_0004.jpg 13
+webcam/images/letter_tray/frame_0005.jpg 13
+webcam/images/letter_tray/frame_0006.jpg 13
+webcam/images/letter_tray/frame_0007.jpg 13
+webcam/images/letter_tray/frame_0008.jpg 13
+webcam/images/letter_tray/frame_0009.jpg 13
+webcam/images/letter_tray/frame_0010.jpg 13
+webcam/images/letter_tray/frame_0011.jpg 13
+webcam/images/letter_tray/frame_0012.jpg 13
+webcam/images/letter_tray/frame_0013.jpg 13
+webcam/images/letter_tray/frame_0014.jpg 13
+webcam/images/letter_tray/frame_0015.jpg 13
+webcam/images/letter_tray/frame_0016.jpg 13
+webcam/images/letter_tray/frame_0017.jpg 13
+webcam/images/letter_tray/frame_0018.jpg 13
+webcam/images/letter_tray/frame_0019.jpg 13
+webcam/images/file_cabinet/frame_0001.jpg 9
+webcam/images/file_cabinet/frame_0002.jpg 9
+webcam/images/file_cabinet/frame_0003.jpg 9
+webcam/images/file_cabinet/frame_0004.jpg 9
+webcam/images/file_cabinet/frame_0005.jpg 9
+webcam/images/file_cabinet/frame_0006.jpg 9
+webcam/images/file_cabinet/frame_0007.jpg 9
+webcam/images/file_cabinet/frame_0008.jpg 9
+webcam/images/file_cabinet/frame_0009.jpg 9
+webcam/images/file_cabinet/frame_0010.jpg 9
+webcam/images/file_cabinet/frame_0011.jpg 9
+webcam/images/file_cabinet/frame_0012.jpg 9
+webcam/images/file_cabinet/frame_0013.jpg 9
+webcam/images/file_cabinet/frame_0014.jpg 9
+webcam/images/file_cabinet/frame_0015.jpg 9
+webcam/images/file_cabinet/frame_0016.jpg 9
+webcam/images/file_cabinet/frame_0017.jpg 9
+webcam/images/file_cabinet/frame_0018.jpg 9
+webcam/images/file_cabinet/frame_0019.jpg 9
+webcam/images/phone/frame_0001.jpg 20
+webcam/images/phone/frame_0002.jpg 20
+webcam/images/phone/frame_0003.jpg 20
+webcam/images/phone/frame_0004.jpg 20
+webcam/images/phone/frame_0005.jpg 20
+webcam/images/phone/frame_0006.jpg 20
+webcam/images/phone/frame_0007.jpg 20
+webcam/images/phone/frame_0008.jpg 20
+webcam/images/phone/frame_0009.jpg 20
+webcam/images/phone/frame_0010.jpg 20
+webcam/images/phone/frame_0011.jpg 20
+webcam/images/phone/frame_0012.jpg 20
+webcam/images/phone/frame_0013.jpg 20
+webcam/images/phone/frame_0014.jpg 20
+webcam/images/phone/frame_0015.jpg 20
+webcam/images/phone/frame_0016.jpg 20
+webcam/images/bookcase/frame_0001.jpg 3
+webcam/images/bookcase/frame_0002.jpg 3
+webcam/images/bookcase/frame_0003.jpg 3
+webcam/images/bookcase/frame_0004.jpg 3
+webcam/images/bookcase/frame_0005.jpg 3
+webcam/images/bookcase/frame_0006.jpg 3
+webcam/images/bookcase/frame_0007.jpg 3
+webcam/images/bookcase/frame_0008.jpg 3
+webcam/images/bookcase/frame_0009.jpg 3
+webcam/images/bookcase/frame_0010.jpg 3
+webcam/images/bookcase/frame_0011.jpg 3
+webcam/images/bookcase/frame_0012.jpg 3
+webcam/images/projector/frame_0001.jpg 22
+webcam/images/projector/frame_0002.jpg 22
+webcam/images/projector/frame_0003.jpg 22
+webcam/images/projector/frame_0004.jpg 22
+webcam/images/projector/frame_0005.jpg 22
+webcam/images/projector/frame_0006.jpg 22
+webcam/images/projector/frame_0007.jpg 22
+webcam/images/projector/frame_0008.jpg 22
+webcam/images/projector/frame_0009.jpg 22
+webcam/images/projector/frame_0010.jpg 22
+webcam/images/projector/frame_0011.jpg 22
+webcam/images/projector/frame_0012.jpg 22
+webcam/images/projector/frame_0013.jpg 22
+webcam/images/projector/frame_0014.jpg 22
+webcam/images/projector/frame_0015.jpg 22
+webcam/images/projector/frame_0016.jpg 22
+webcam/images/projector/frame_0017.jpg 22
+webcam/images/projector/frame_0018.jpg 22
+webcam/images/projector/frame_0019.jpg 22
+webcam/images/projector/frame_0020.jpg 22
+webcam/images/projector/frame_0021.jpg 22
+webcam/images/projector/frame_0022.jpg 22
+webcam/images/projector/frame_0023.jpg 22
+webcam/images/projector/frame_0024.jpg 22
+webcam/images/projector/frame_0025.jpg 22
+webcam/images/projector/frame_0026.jpg 22
+webcam/images/projector/frame_0027.jpg 22
+webcam/images/projector/frame_0028.jpg 22
+webcam/images/projector/frame_0029.jpg 22
+webcam/images/projector/frame_0030.jpg 22
+webcam/images/stapler/frame_0001.jpg 28
+webcam/images/stapler/frame_0002.jpg 28
+webcam/images/stapler/frame_0003.jpg 28
+webcam/images/stapler/frame_0004.jpg 28
+webcam/images/stapler/frame_0005.jpg 28
+webcam/images/stapler/frame_0006.jpg 28
+webcam/images/stapler/frame_0007.jpg 28
+webcam/images/stapler/frame_0008.jpg 28
+webcam/images/stapler/frame_0009.jpg 28
+webcam/images/stapler/frame_0010.jpg 28
+webcam/images/stapler/frame_0011.jpg 28
+webcam/images/stapler/frame_0012.jpg 28
+webcam/images/stapler/frame_0013.jpg 28
+webcam/images/stapler/frame_0014.jpg 28
+webcam/images/stapler/frame_0015.jpg 28
+webcam/images/stapler/frame_0016.jpg 28
+webcam/images/stapler/frame_0017.jpg 28
+webcam/images/stapler/frame_0018.jpg 28
+webcam/images/stapler/frame_0019.jpg 28
+webcam/images/stapler/frame_0020.jpg 28
+webcam/images/stapler/frame_0021.jpg 28
+webcam/images/stapler/frame_0022.jpg 28
+webcam/images/stapler/frame_0023.jpg 28
+webcam/images/stapler/frame_0024.jpg 28
+webcam/images/trash_can/frame_0001.jpg 30
+webcam/images/trash_can/frame_0002.jpg 30
+webcam/images/trash_can/frame_0003.jpg 30
+webcam/images/trash_can/frame_0004.jpg 30
+webcam/images/trash_can/frame_0005.jpg 30
+webcam/images/trash_can/frame_0006.jpg 30
+webcam/images/trash_can/frame_0007.jpg 30
+webcam/images/trash_can/frame_0008.jpg 30
+webcam/images/trash_can/frame_0009.jpg 30
+webcam/images/trash_can/frame_0010.jpg 30
+webcam/images/trash_can/frame_0011.jpg 30
+webcam/images/trash_can/frame_0012.jpg 30
+webcam/images/trash_can/frame_0013.jpg 30
+webcam/images/trash_can/frame_0014.jpg 30
+webcam/images/trash_can/frame_0015.jpg 30
+webcam/images/trash_can/frame_0016.jpg 30
+webcam/images/trash_can/frame_0017.jpg 30
+webcam/images/trash_can/frame_0018.jpg 30
+webcam/images/trash_can/frame_0019.jpg 30
+webcam/images/trash_can/frame_0020.jpg 30
+webcam/images/trash_can/frame_0021.jpg 30
+webcam/images/bike_helmet/frame_0001.jpg 2
+webcam/images/bike_helmet/frame_0002.jpg 2
+webcam/images/bike_helmet/frame_0003.jpg 2
+webcam/images/bike_helmet/frame_0004.jpg 2
+webcam/images/bike_helmet/frame_0005.jpg 2
+webcam/images/bike_helmet/frame_0006.jpg 2
+webcam/images/bike_helmet/frame_0007.jpg 2
+webcam/images/bike_helmet/frame_0008.jpg 2
+webcam/images/bike_helmet/frame_0009.jpg 2
+webcam/images/bike_helmet/frame_0010.jpg 2
+webcam/images/bike_helmet/frame_0011.jpg 2
+webcam/images/bike_helmet/frame_0012.jpg 2
+webcam/images/bike_helmet/frame_0013.jpg 2
+webcam/images/bike_helmet/frame_0014.jpg 2
+webcam/images/bike_helmet/frame_0015.jpg 2
+webcam/images/bike_helmet/frame_0016.jpg 2
+webcam/images/bike_helmet/frame_0017.jpg 2
+webcam/images/bike_helmet/frame_0018.jpg 2
+webcam/images/bike_helmet/frame_0019.jpg 2
+webcam/images/bike_helmet/frame_0020.jpg 2
+webcam/images/bike_helmet/frame_0021.jpg 2
+webcam/images/bike_helmet/frame_0022.jpg 2
+webcam/images/bike_helmet/frame_0023.jpg 2
+webcam/images/bike_helmet/frame_0024.jpg 2
+webcam/images/bike_helmet/frame_0025.jpg 2
+webcam/images/bike_helmet/frame_0026.jpg 2
+webcam/images/bike_helmet/frame_0027.jpg 2
+webcam/images/bike_helmet/frame_0028.jpg 2
+webcam/images/headphones/frame_0001.jpg 10
+webcam/images/headphones/frame_0002.jpg 10
+webcam/images/headphones/frame_0003.jpg 10
+webcam/images/headphones/frame_0004.jpg 10
+webcam/images/headphones/frame_0005.jpg 10
+webcam/images/headphones/frame_0006.jpg 10
+webcam/images/headphones/frame_0007.jpg 10
+webcam/images/headphones/frame_0008.jpg 10
+webcam/images/headphones/frame_0009.jpg 10
+webcam/images/headphones/frame_0010.jpg 10
+webcam/images/headphones/frame_0011.jpg 10
+webcam/images/headphones/frame_0012.jpg 10
+webcam/images/headphones/frame_0013.jpg 10
+webcam/images/headphones/frame_0014.jpg 10
+webcam/images/headphones/frame_0015.jpg 10
+webcam/images/headphones/frame_0016.jpg 10
+webcam/images/headphones/frame_0017.jpg 10
+webcam/images/headphones/frame_0018.jpg 10
+webcam/images/headphones/frame_0019.jpg 10
+webcam/images/headphones/frame_0020.jpg 10
+webcam/images/headphones/frame_0021.jpg 10
+webcam/images/headphones/frame_0022.jpg 10
+webcam/images/headphones/frame_0023.jpg 10
+webcam/images/headphones/frame_0024.jpg 10
+webcam/images/headphones/frame_0025.jpg 10
+webcam/images/headphones/frame_0026.jpg 10
+webcam/images/headphones/frame_0027.jpg 10
+webcam/images/desk_lamp/frame_0001.jpg 7
+webcam/images/desk_lamp/frame_0002.jpg 7
+webcam/images/desk_lamp/frame_0003.jpg 7
+webcam/images/desk_lamp/frame_0004.jpg 7
+webcam/images/desk_lamp/frame_0005.jpg 7
+webcam/images/desk_lamp/frame_0006.jpg 7
+webcam/images/desk_lamp/frame_0007.jpg 7
+webcam/images/desk_lamp/frame_0008.jpg 7
+webcam/images/desk_lamp/frame_0009.jpg 7
+webcam/images/desk_lamp/frame_0010.jpg 7
+webcam/images/desk_lamp/frame_0011.jpg 7
+webcam/images/desk_lamp/frame_0012.jpg 7
+webcam/images/desk_lamp/frame_0013.jpg 7
+webcam/images/desk_lamp/frame_0014.jpg 7
+webcam/images/desk_lamp/frame_0015.jpg 7
+webcam/images/desk_lamp/frame_0016.jpg 7
+webcam/images/desk_lamp/frame_0017.jpg 7
+webcam/images/desk_lamp/frame_0018.jpg 7
+webcam/images/desk_chair/frame_0001.jpg 6
+webcam/images/desk_chair/frame_0002.jpg 6
+webcam/images/desk_chair/frame_0003.jpg 6
+webcam/images/desk_chair/frame_0004.jpg 6
+webcam/images/desk_chair/frame_0005.jpg 6
+webcam/images/desk_chair/frame_0006.jpg 6
+webcam/images/desk_chair/frame_0007.jpg 6
+webcam/images/desk_chair/frame_0008.jpg 6
+webcam/images/desk_chair/frame_0009.jpg 6
+webcam/images/desk_chair/frame_0010.jpg 6
+webcam/images/desk_chair/frame_0011.jpg 6
+webcam/images/desk_chair/frame_0012.jpg 6
+webcam/images/desk_chair/frame_0013.jpg 6
+webcam/images/desk_chair/frame_0014.jpg 6
+webcam/images/desk_chair/frame_0015.jpg 6
+webcam/images/desk_chair/frame_0016.jpg 6
+webcam/images/desk_chair/frame_0017.jpg 6
+webcam/images/desk_chair/frame_0018.jpg 6
+webcam/images/desk_chair/frame_0019.jpg 6
+webcam/images/desk_chair/frame_0020.jpg 6
+webcam/images/desk_chair/frame_0021.jpg 6
+webcam/images/desk_chair/frame_0022.jpg 6
+webcam/images/desk_chair/frame_0023.jpg 6
+webcam/images/desk_chair/frame_0024.jpg 6
+webcam/images/desk_chair/frame_0025.jpg 6
+webcam/images/desk_chair/frame_0026.jpg 6
+webcam/images/desk_chair/frame_0027.jpg 6
+webcam/images/desk_chair/frame_0028.jpg 6
+webcam/images/desk_chair/frame_0029.jpg 6
+webcam/images/desk_chair/frame_0030.jpg 6
+webcam/images/desk_chair/frame_0031.jpg 6
+webcam/images/desk_chair/frame_0032.jpg 6
+webcam/images/desk_chair/frame_0033.jpg 6
+webcam/images/desk_chair/frame_0034.jpg 6
+webcam/images/desk_chair/frame_0035.jpg 6
+webcam/images/desk_chair/frame_0036.jpg 6
+webcam/images/desk_chair/frame_0037.jpg 6
+webcam/images/desk_chair/frame_0038.jpg 6
+webcam/images/desk_chair/frame_0039.jpg 6
+webcam/images/desk_chair/frame_0040.jpg 6
+webcam/images/bottle/frame_0001.jpg 4
+webcam/images/bottle/frame_0002.jpg 4
+webcam/images/bottle/frame_0003.jpg 4
+webcam/images/bottle/frame_0004.jpg 4
+webcam/images/bottle/frame_0005.jpg 4
+webcam/images/bottle/frame_0006.jpg 4
+webcam/images/bottle/frame_0007.jpg 4
+webcam/images/bottle/frame_0008.jpg 4
+webcam/images/bottle/frame_0009.jpg 4
+webcam/images/bottle/frame_0010.jpg 4
+webcam/images/bottle/frame_0011.jpg 4
+webcam/images/bottle/frame_0012.jpg 4
+webcam/images/bottle/frame_0013.jpg 4
+webcam/images/bottle/frame_0014.jpg 4
+webcam/images/bottle/frame_0015.jpg 4
+webcam/images/bottle/frame_0016.jpg 4
--- a/data/office-home/Art.txt
+++ b/data/office-home/Art.txt
--- a/data/office-home/Clipart.txt
+++ b/data/office-home/Clipart.txt
--- a/data/office-home/Product.txt
+++ b/data/office-home/Product.txt
--- a/data/office-home/Real_World.txt
+++ b/data/office-home/Real_World.txt
--- a/data/usps2mnist/mnist_test.txt
+++ b/data/usps2mnist/mnist_test.txt
--- a/data/usps2mnist/mnist_train.txt
+++ b/data/usps2mnist/mnist_train.txt
--- a/data/usps2mnist/usps_test.txt
+++ b/data/usps2mnist/usps_test.txt
--- a/data/usps2mnist/usps_train.txt
+++ b/data/usps2mnist/usps_train.txt
--- a/data/visda/train_list.txt
+++ b/data/visda/train_list.txt
--- a/data/visda/validation_list.txt
+++ b/data/visda/validation_list.txt
--- a/data_list.py
+++ b/data_list.py
@ -0,0 +1,234 @@
+#from __future__ import print_function, division
+
+import scipy.stats
+import numpy as np
+import os
+import os.path
+import pickle
+import random
+import sys
+import time
+import torch
+from PIL import Image
+from torch.utils.data import Dataset
+import torch.nn as nn
+from torchvision import datasets, transforms
+
+
+def write_list(f, l):
+    f.write(",".join(map(str, l)) + "\n")
+    f.flush()
+
+
+def sample_ratios(samples, labels, ratios=None):
+    if ratios is not None:
+        selected_idx = []
+        for i, ratio in enumerate(ratios):
+            images_i = [j for j in range(
+                labels.shape[0]) if labels[j].item() == i]
+            num_ = len(images_i)
+            idx = np.random.choice(
+                num_, int(ratio * num_), replace=False)
+            selected_idx.extend(np.array(images_i)[idx].tolist())
+        return samples[selected_idx, :, :], labels[selected_idx]
+
+    return samples, labels
+
+
+def image_classification_test_loaded(test_samples, test_labels, model, test_10crop=True, device='cpu'):
+    with torch.no_grad():
+        test_loss = 0
+        correct = 0
+        if test_10crop:
+            len_test = test_labels[0].shape[0]
+            for i in range(len_test):
+                outputs = []
+                for j in range(10):
+                    data, target = test_samples[j][i].unsqueeze(
+                        0), test_labels[j][i].unsqueeze(0)
+                    _, output = model(data)
+                    test_loss += nn.CrossEntropyLoss()(output, target).item()
+                    outputs.append(nn.Softmax(dim=1)(output))
+                outputs = sum(outputs)
+                pred = torch.max(outputs, 1)[1]
+                correct += pred.eq(target.data.cpu().view_as(pred)
+                                   ).sum().item()
+        else:
+            len_test = test_labels.shape[0]
+            bs = 72
+            for i in range(int(len_test / bs)):
+                data, target = torch.Tensor(
+                    test_samples[bs*i:bs*(i+1)]).to(device), test_labels[bs*i:bs*(i+1)]
+                output = model(data)
+                test_loss += nn.CrossEntropyLoss()(output, target).item()
+                pred = torch.max(output, 1)[1]
+                correct += pred.eq(target.data.view_as(pred)).sum().item()
+            # Last test samples
+            data, target = torch.Tensor(
+                test_samples[bs*(i+1):]).to(device), test_labels[bs*(i+1):]
+            output = model(data)
+            test_loss += nn.CrossEntropyLoss()(output, target).item()
+            pred = torch.max(output, 1)[1]
+            correct += pred.eq(target.data.view_as(pred)).sum().item()
+    accuracy = correct / len_test
+    test_loss /= len_test
+    return accuracy
+
+
+def make_dataset(image_list, labels, ratios=None):
+    if labels:
+      len_ = len(image_list)
+      images = [(image_list[i].strip(), labels[i, :]) for i in range(len_)]
+    else:
+      if len(image_list[0].split()) > 2:
+        images = [(val.split()[0], np.array([int(la) for la in val.split()[1:]])) for val in image_list]
+      else:
+        images = [(val.split()[0], int(val.split()[1])) for val in image_list]
+    if ratios:
+        selected_images = []
+        for i, ratio in enumerate(ratios):
+            images_i = [img for img in images if img[1] == i]
+            num_ = len(images_i)
+            idx = np.random.choice(num_, int(ratio * num_), replace=False)
+            for j in idx:
+                selected_images.append(images_i[j])
+        return selected_images
+
+    return images
+
+
+def rgb_loader(path):
+    with open(path, 'rb') as f:
+        with Image.open(f) as img:
+            return img.convert('RGB')
+
+
+def l_loader(path):
+    with open(path, 'rb') as f:
+        with Image.open(f) as img:
+            return img.convert('L')
+
+
+def build_uspsmnist(l, path, root_folder, device='cpu'):
+    dset_source = ImageList(open(l).readlines(), transform=transforms.Compose([
+        transforms.Resize((28, 28)),
+        transforms.ToTensor(),
+        transforms.Normalize((0.5,), (0.5,))
+    ]), mode='L', root_folder=root_folder)
+    loaded_dset_source = LoadedImageList(dset_source)
+    with open(path, 'wb') as f:
+        pickle.dump([loaded_dset_source.samples.numpy(),
+                     loaded_dset_source.targets.numpy()], f)
+    return loaded_dset_source.samples.to(device
+                                         ), loaded_dset_source.targets.to(device)
+
+
+class ImageList(Dataset):
+    def __init__(self, image_list, labels=None, transform=None, target_transform=None, mode='RGB', root_folder='', ratios=None):
+        imgs = make_dataset(image_list, labels, ratios=ratios)
+        if len(imgs) == 0:
+            raise(RuntimeError("Found 0 images in subfolders of: " + root_folder + "\n"))
+
+        self.root_folder = root_folder
+        self.imgs = imgs
+        self.transform = transform
+        self.target_transform = target_transform
+        if mode == 'RGB':
+            self.loader = rgb_loader
+        elif mode == 'L':
+            self.loader = l_loader
+
+    def __getitem__(self, index):
+        path, target = self.imgs[index]
+        img = self.loader(os.path.join(self.root_folder, path))
+        if self.transform is not None:
+            img = self.transform(img)
+        if self.target_transform is not None:
+            target = self.target_transform(target)
+        return img, target
+
+    def __len__(self):
+        return len(self.imgs)
+
+
+class LoadedImageList(Dataset):
+    def __init__(self, image_list):
+
+        self.image_list = image_list
+        self.samples, self.targets = self._load_imgs()
+
+    def _load_imgs(self):
+
+        loaded_images, targets = [], []
+        t = time.time()
+        print("{} samples to process".format(len(self.image_list.imgs)))
+        for i, (path, target) in enumerate(self.image_list.imgs):
+            if i % 1000 == 999:
+                print("{} samples in {} seconds".format(i, time.time()- t))
+                sys.stdout.flush()
+            img = self.image_list.loader(os.path.join(self.image_list.root_folder, path))
+            if self.image_list.transform is not None:
+                img = self.image_list.transform(img)
+            if self.image_list.target_transform is not None:
+                target = self.image_list.target_transform(target)
+            loaded_images.append(img)
+            targets.append(target)
+
+        return torch.stack(loaded_images), torch.LongTensor(targets)
+
+    def __len__(self):
+        return len(self.image_list.imgs)
+
+
+# List of fractions used to produce the dataset in the Performance vs J_SD paragraph. Figures 1 and 2.
+# 50 elements, representing the fractions, and the theoretical JSD they would produce.
+subsampling = [[[0.4052, 0.2398, 0.7001, 0.7178, 0.2132, 0.4887, 0.9849, 0.814, 0.8186, 0.2134], 0.032760409197316966],
+                [[0.7872, 0.6374, 0.8413, 0.3612, 0.4427, 0.5154, 0.8423, 0.6748, 0.1634, 0.4098], 0.021267958321062656],
+                [[0.2789, 0.5613, 0.9585, 0.2165, 0.9446, 0.869, 0.838, 0.9575, 0.1829, 0.7934], 0.03455597584383796],
+                [[0.5017, 0.4318, 0.6189, 0.1269, 0.8788, 0.4277, 0.3875, 0.5378, 0.5405, 0.8286], 0.022225051878141944],
+                [[0.2642, 0.1912, 0.1133, 0.7818, 0.9773, 0.9555, 0.741, 0.6051, 0.2096, 0.8654], 0.05013218434133124],
+                [[0.2005, 0.579, 0.5852, 0.814, 0.3936, 0.6541, 0.6803, 0.3873, 0.5198, 0.7423], 0.01487691279068412],
+                [[0.6315, 0.9692, 0.5519, 0.317, 0.662, 0.9937, 0.577, 0.895, 0.2807, 0.5573], 0.01786388435441286],
+                [[0.6593, 0.1823, 0.9087, 0.323, 0.117, 0.324, 0.9852, 0.8075, 0.9799, 0.1527], 0.05721295477923642],
+                [[0.9851, 0.5631, 0.3203, 0.227, 0.5126, 0.3978, 0.5219, 0.9644, 0.427, 0.8089], 0.02346584399291991],
+                [[0.9434, 0.4466, 0.8503, 0.7235, 0.405, 0.8225, 0.2699, 0.6799, 0.5503, 0.8828], 0.01554582865827399],
+                [[0.7742, 0.9749, 0.895, 0.3821, 0.7929, 0.5722, 0.5327, 0.4774, 0.7049, 0.1087], 0.026333395357855234],
+                [[0.474, 0.4977, 0.4882, 0.7236, 0.4274, 0.505, 0.1939, 0.6071, 0.7333, 0.6566], 0.012314910203020654],
+                [[0.7469, 0.1428, 0.7743, 0.261, 0.509, 0.6714, 0.3558, 0.2003, 0.2123, 0.7144], 0.03758103804746866],
+                [[0.6078, 0.9925, 0.6132, 0.5107, 0.7755, 0.266, 0.9181, 0.5503, 0.9753, 0.787], 0.014173062912087135],
+                [[0.4534, 0.3197, 0.641, 0.4995, 0.8052, 0.5237, 0.6757, 0.6292, 0.8155, 0.8259], 0.009193941950225593],
+                [[0.3339, 0.7985, 0.3739, 0.8049, 0.4443, 0.5783, 0.6133, 0.7278, 0.9127, 0.7342], 0.011946408195133998],
+                [[0.513, 0.6891, 0.8503, 0.7699, 0.8338, 0.8842, 0.2925, 0.5097, 0.8512, 0.9673], 0.011943243518739478],
+                [[0.2608, 0.6524, 0.8783, 0.5355, 0.8859, 0.325, 0.8223, 0.2732, 0.3384, 0.733], 0.024573748979216877],
+                [[0.4179, 0.8885, 0.778, 0.5943, 0.4063, 0.7704, 0.3782, 0.3344, 0.784, 0.7756], 0.01417171679861576],
+                [[0.721, 0.4303, 0.5113, 0.8499, 0.6216, 0.3463, 0.7363, 0.438, 0.2705, 0.6982], 0.013848046051355988],
+                [[0.9136, 0.964, 0.9482, 0.7971, 0.4281, 0.4715, 0.8894, 0.272, 0.3951, 0.7778], 0.019194643091215248],
+                [[0.4694, 0.5974, 0.6888, 0.7073, 0.5244, 0.5828, 0.4859, 0.8798, 0.6837, 0.3665], 0.006838157001020787],
+                [[0.5397, 0.8727, 0.3284, 0.781, 0.3955, 0.88, 0.3357, 0.8478, 0.8832, 0.4495], 0.017877156258920744],
+                [[0.4096, 0.6769, 0.3566, 0.3863, 0.6231, 0.7828, 0.5524, 0.5988, 0.8546, 0.3649], 0.011476193912275862],
+                [[0.6329, 0.5326, 0.8094, 0.556, 0.2853, 0.2693, 0.4511, 0.6068, 0.4359, 0.7832], 0.014025816871972494],
+                [[0.6955, 0.9107, 0.9266, 0.7358, 0.7268, 0.5914, 0.7524, 0.9486, 0.6223, 0.92], 0.003294516531428644],
+                [[0.6221, 0.5749, 0.4356, 0.2805, 0.3879, 0.8183, 0.9777, 0.3194, 0.6435, 0.7227], 0.017521988550845864],
+                [[0.7197, 0.4776, 0.9488, 0.9869, 0.6423, 0.3082, 0.4404, 0.7307, 0.9051, 0.5027], 0.014615851037701265],
+                [[0.7597, 0.9354, 0.6758, 0.9101, 0.8474, 0.6598, 0.7983, 0.6623, 0.8925, 0.7054], 0.0020994981073005157],
+                [[0.8551, 0.6381, 0.6828, 0.8635, 0.6717, 0.6722, 0.8073, 0.7905, 0.9169, 0.7565], 0.0017818497157099332],
+                [[0.7233, 0.7575, 0.8205, 0.6374, 0.6596, 0.7792, 0.8736, 0.8413, 0.807, 0.7333], 0.0011553039532261685],
+                [[0.8778, 0.9661, 0.6376, 0.9134, 0.8296, 0.7027, 0.977, 0.8485, 0.6357, 0.7192], 0.0029014496151538102],
+                [[0.9774, 0.9047, 0.9077, 0.9301, 0.9926, 0.93, 0.9202, 0.9849, 0.9231, 0.9612], 0.0001353783157801754],
+                [[0.6304, 0.1495, 0.8439, 0.3944, 0.8693, 0.1466, 0.5952, 0.5705, 0.5734, 0.9339], 0.0337553984425215],
+                [[0.6239, 0.9361, 0.7968, 0.7829, 0.7501, 0.1326, 0.6832, 0.1769, 0.616, 0.319], 0.033912799130997845],
+                [[0.1608, 0.1367, 0.797, 0.655, 0.3387, 0.539, 0.4643, 0.6889, 0.3485, 0.1786], 0.037503662696288936],
+                [[0.7596, 0.6143, 0.1171, 0.4069, 0.5707, 0.2775, 0.6583, 0.1072, 0.9349, 0.3256], 0.044312727351805296],
+                [[0.1518, 0.9251, 0.2583, 0.5876, 0.4173, 0.9602, 0.1169, 0.5076, 0.9346, 0.8579], 0.04662314442412397],
+                [[0.1608, 0.9229, 0.3684, 0.9401, 0.2972, 0.8121, 0.332, 0.1203, 0.8746, 0.6694], 0.046668910811104636],
+                [[0.9124, 0.7087, 0.5254, 0.3416, 0.6167, 0.243, 0.3906, 0.1258, 0.1402, 0.749], 0.04128706833470013],
+                [[0.4195, 0.778, 0.2067, 0.7385, 0.2988, 0.1888, 0.1895, 0.2513, 0.7798, 0.2208], 0.041400745619836664],
+                [[0.2199, 0.4556, 0.6841, 0.3873, 0.9961, 0.1154, 0.9678, 0.7335, 0.1152, 0.8384], 0.05217544485450189],
+                [[0.2801, 0.2895, 0.2793, 0.83, 0.4832, 0.9456, 0.139, 0.402, 0.9296, 0.1218], 0.051751059034301244],
+                [[0.1197, 0.2005, 0.5692, 0.1897, 0.9014, 0.2501, 0.6541, 0.2476, 0.9808, 0.9541], 0.056797377461948295],
+                [[0.17, 0.1027, 0.5433, 0.9669, 0.1256, 0.6906, 0.6202, 0.3288, 0.7465, 0.738], 0.050916190636527255],
+                [[0.893, 0.4175, 0.1314, 0.1503, 0.7415, 0.5234, 0.1702, 0.7846, 0.7492, 0.1314], 0.056894089148603375],
+                [[0.8962, 0.1078, 0.951, 0.8255, 0.1396, 0.129, 0.2559, 0.7041, 0.9513, 0.9961], 0.06364930131498307],
+                [[0.1053, 0.4421, 0.1935, 0.1014, 0.5312, 0.7668, 0.3904, 0.958, 0.1548, 0.7625], 0.061061160204041176],
+                [[0.11, 0.2594, 0.6996, 0.7597, 0.1755, 0.1002, 0.1926, 0.4621, 0.9265, 0.8734], 0.06539390880498576],
+                [[0.927, 0.9604, 0.1183, 0.3875, 0.134, 0.7651, 0.8139, 0.1396, 0.5639, 0.1171], 0.06938724191363951]]
--- a/loss.py
+++ b/loss.py
@ -0,0 +1,245 @@
+import numpy as np
+import torch
+import torch.nn as nn
+from torch.autograd import Variable
+import math
+import torch.nn.functional as F
+import pdb
+
+
+def Entropy(input_):
+    bs = input_.size(0)
+    epsilon = 1e-5
+    entropy = -input_ * torch.log(input_ + epsilon)
+    entropy = torch.sum(entropy, dim=1)
+    return entropy
+
+
+def grl_hook(coeff):
+    def fun1(grad):
+        return -coeff*grad.clone()
+    return fun1
+
+
+def CDAN(input_list, ad_net, entropy=None, coeff=None, random_layer=None, weights=None, device='cuda'):
+    softmax_output = input_list[1].detach()
+    batch_size = softmax_output.size(0) // 2
+    feature = input_list[0]
+    if random_layer is None:
+        op_out = torch.bmm(softmax_output.unsqueeze(2), feature.unsqueeze(1))
+        ad_out = ad_net(
+            op_out.view(-1, softmax_output.size(1) * feature.size(1)))
+    else:
+        random_out = random_layer.forward([feature, softmax_output])
+        ad_out = ad_net(random_out.view(-1, random_out.size(1)))
+    dc_target = torch.from_numpy(
+        np.array([[1]] * batch_size + [[0]] * batch_size)).float().to(device)
+    if entropy is not None:
+        entropy.register_hook(grl_hook(coeff))
+        entropy = 1.0+torch.exp(-entropy)
+        source_mask = torch.ones_like(entropy)
+        source_mask[feature.size(0)//2:] = 0
+        source_weight = entropy*source_mask
+        target_mask = torch.ones_like(entropy)
+        target_mask[0:feature.size(0)//2] = 0
+        target_weight = entropy*target_mask
+        if weights is not None:
+            weights = torch.cat((weights.squeeze(), torch.ones((batch_size)).to(device)))
+            weight = (source_weight / torch.sum(source_weight).detach().item() +
+                    target_weight / torch.sum(target_weight).detach().item()) * weights
+        else:
+            weight = source_weight / torch.sum(source_weight).detach().item() + \
+                     target_weight / torch.sum(target_weight).detach().item()
+        return torch.sum(weight.view(-1, 1) * nn.BCELoss(reduction='none')(ad_out, dc_target)) / torch.sum(weight).detach().item()
+    else:
+        if weights is not None:
+            weighted_nll_source = - weights * torch.log(ad_out[:batch_size])
+            nll_target = - torch.log(1 - ad_out[batch_size:])
+            return (torch.mean(weighted_nll_source) + torch.mean(nll_target)) / 2
+
+        return nn.BCELoss()(ad_out, dc_target)
+
+
+def DANN(features, ad_net, device):
+    ad_out = ad_net(features)
+    batch_size = ad_out.size(0) // 2
+    dc_target = torch.from_numpy(
+        np.array([[1]] * batch_size + [[0]] * batch_size)).float().to(device)
+    return nn.BCELoss()(ad_out, dc_target)
+
+
+def IWDAN(features, ad_net, weights):
+
+    # First batch_size elements of features correspond to source
+    # Last  batch_size elements of features correspond to target
+    # Each element of ad_out represents the proba of the corresponding feature to be from the source domain
+    # For importance sampling, it needs to be put to the log and multiplied by the weight of the corresponding class
+
+    ad_out = ad_net(features)
+    batch_size = ad_out.size(0) // 2
+
+    weighted_nll_source = - weights * torch.log(ad_out[:batch_size])
+    nll_target = - torch.log(1 - ad_out[batch_size:])
+
+    return (torch.mean(weighted_nll_source) + torch.mean(nll_target)) / 2
+
+
+def WDANN(features, ad_net, device, weights=None):
+    ad_out = ad_net(features)
+    batch_size = ad_out.size(0) // 2
+    if weights is None:
+        weighted_source = ad_out[:batch_size]
+    else:
+        weighted_source = ad_out[:batch_size] * weights
+    dc_target = torch.from_numpy(
+        np.array([[1]] * batch_size + [[-1]] * batch_size)).float().to(device)
+
+    # Gradient penalty
+    alpha = torch.rand([batch_size, 1]).to(device)
+    interpolates = (1 - alpha) * features[batch_size:] + alpha * features[:batch_size]
+    interpolates = torch.autograd.Variable(interpolates, requires_grad=True)
+    disc_interpolates = ad_net(interpolates)
+    gradients = torch.autograd.grad(outputs=disc_interpolates, inputs=interpolates,
+                                    grad_outputs=torch.ones(disc_interpolates.size()).to(device),
+                                    create_graph=True, retain_graph=True, only_inputs=True)[0]
+    gradients = gradients.view(gradients.size(0), -1)
+
+    # Lambda is 10 in the original WGAN-GP paper
+    # return - torch.mean(dc_target * ad_out), ((gradients.norm(2, dim=1) - 1) ** 2).mean() * 10
+    return - torch.mean(weighted_source - ad_out[batch_size:]) / 2, ((gradients.norm(2, dim=1) - 1) ** 2).mean() * 10
+
+
+def EntropyLoss(input_):
+    mask = input_.ge(0.000001)
+    mask_out = torch.masked_select(input_, mask)
+    entropy = -(torch.sum(mask_out * torch.log(mask_out)))
+    return entropy / float(input_.size(0))
+
+
+def gaussian_kernel(source, target, kernel_mul=2.0, kernel_num=5, fix_sigma=None):
+    n_samples = int(source.size()[0])+int(target.size()[0])
+    total = torch.cat([source, target], dim=0)
+    total0 = total.unsqueeze(0).expand(
+        int(total.size(0)), int(total.size(0)), int(total.size(1)))
+    total1 = total.unsqueeze(1).expand(
+        int(total.size(0)), int(total.size(0)), int(total.size(1)))
+    L2_distance = ((total0-total1)**2).sum(2)
+    if fix_sigma:
+        bandwidth = fix_sigma
+    else:
+        bandwidth = torch.sum(L2_distance.data) / (n_samples**2-n_samples)
+    bandwidth /= kernel_mul ** (kernel_num // 2)
+    bandwidth_list = [bandwidth * (kernel_mul**i) for i in range(kernel_num)]
+    kernel_val = [torch.exp(-L2_distance / bandwidth_temp)
+                  for bandwidth_temp in bandwidth_list]
+    return sum(kernel_val)  # /len(kernel_val)
+
+
+def DAN(source, target, kernel_mul=2.0, kernel_num=5, fix_sigma=None):
+    batch_size = int(source.size()[0])
+    kernels = gaussian_kernel(source, target,
+                              kernel_mul=kernel_mul, kernel_num=kernel_num, fix_sigma=fix_sigma)
+
+    loss1 = 0
+    for s1 in range(batch_size):
+        for s2 in range(s1+1, batch_size):
+            t1, t2 = s1+batch_size, s2+batch_size
+            loss1 += kernels[s1, s2] + kernels[t1, t2]
+    loss1 = loss1 / float(batch_size * (batch_size - 1) / 2)
+
+    loss2 = 0
+    for s1 in range(batch_size):
+        for s2 in range(batch_size):
+            t1, t2 = s1+batch_size, s2+batch_size
+            loss2 -= kernels[s1, t2] + kernels[s2, t1]
+    loss2 = loss2 / float(batch_size * batch_size)
+    return loss1 + loss2
+
+
+def DAN_Linear(source, target, kernel_mul=2.0, kernel_num=5, fix_sigma=None):
+    batch_size = int(source.size()[0])
+    kernels = gaussian_kernel(source, target,
+                              kernel_mul=kernel_mul, kernel_num=kernel_num, fix_sigma=fix_sigma)
+
+    # Linear version
+    loss = 0
+    for i in range(batch_size):
+        s1, s2 = i, (i+1) % batch_size
+        t1, t2 = s1+batch_size, s2+batch_size
+        loss += kernels[s1, s2] + kernels[t1, t2]
+        loss -= kernels[s1, t2] + kernels[s2, t1]
+    return loss / float(batch_size)
+
+
+def JAN(source_list, target_list, kernel_muls=[2.0, 2.0], kernel_nums=[5, 1], fix_sigma_list=[None, 1.68], weights=None):
+    batch_size = int(source_list[0].size()[0])
+    layer_num = len(source_list)
+    joint_kernels = None
+    for i in range(layer_num):
+        source = source_list[i]
+        target = target_list[i]
+        kernel_mul = kernel_muls[i]
+        kernel_num = kernel_nums[i]
+        fix_sigma = fix_sigma_list[i]
+        kernels = gaussian_kernel(source, target,
+                                  kernel_mul=kernel_mul, kernel_num=kernel_num, fix_sigma=fix_sigma)
+        if joint_kernels is not None:
+            joint_kernels = joint_kernels * kernels
+        else:
+            joint_kernels = kernels
+
+    loss1 = 0
+    if weights is None:
+        mult = 1
+    for s1 in range(batch_size):
+        for s2 in range(s1 + 1, batch_size):
+            t1, t2 = s1 + batch_size, s2 + batch_size
+            if weights is not None:
+                mult = weights[s1] * weights[s2]
+            loss1 += mult * joint_kernels[s1, s2] + joint_kernels[t1, t2]
+    loss1 = loss1 / float(batch_size * (batch_size - 1) / 2)
+
+    loss2 = 0
+    if weights is None:
+        mult1, mult2 = 1, 1
+    for s1 in range(batch_size):
+        if weights is not None:
+            mult1 = weights[s1]
+        for s2 in range(batch_size):
+            t1, t2 = s1 + batch_size, s2 + batch_size
+            if weights is not None:
+                mult2 = weights[s2]
+            loss2 -= mult1 * joint_kernels[s1, t2] + mult2 * joint_kernels[s2, t1]
+    loss2 = loss2 / float(batch_size * batch_size)
+    return loss1 + loss2
+
+
+def JAN_Linear(source_list, target_list, kernel_muls=[2.0, 2.0], kernel_nums=[5, 1], fix_sigma_list=[None, 1.68]):
+    batch_size = int(source_list[0].size()[0])
+    layer_num = len(source_list)
+    joint_kernels = None
+    for i in range(layer_num):
+        source = source_list[i]
+        target = target_list[i]
+        kernel_mul = kernel_muls[i]
+        kernel_num = kernel_nums[i]
+        fix_sigma = fix_sigma_list[i]
+        kernels = gaussian_kernel(source, target,
+                                  kernel_mul=kernel_mul, kernel_num=kernel_num, fix_sigma=fix_sigma)
+        if joint_kernels is not None:
+            joint_kernels = joint_kernels * kernels
+        else:
+            joint_kernels = kernels
+
+    # Linear version
+    loss = 0
+    for i in range(batch_size):
+        s1, s2 = i, (i+1) % batch_size
+        t1, t2 = s1+batch_size, s2+batch_size
+        loss += joint_kernels[s1, s2] + joint_kernels[t1, t2]
+        loss -= joint_kernels[s1, t2] + joint_kernels[s2, t1]
+    return loss / float(batch_size)
+
+
+loss_dict = {"DAN": DAN, "DAN_Linear": DAN_Linear, "JAN": JAN,
+             "JAN_Linear": JAN_Linear, "IWJAN": JAN, "IWJANORACLE": JAN}
--- a/lr_schedule.py
+++ b/lr_schedule.py
@ -0,0 +1,24 @@
+def inv_lr_scheduler(optimizer, iter_num, gamma, power, lr=0.001, weight_decay=0.0005):
+    """Decay learning rate by a factor of 0.1 every lr_decay_epoch epochs."""
+    lr = lr * (1 + gamma * iter_num) ** (-power)
+    i=0
+    for param_group in optimizer.param_groups:
+        param_group['lr'] = lr * param_group['lr_mult']
+        param_group['weight_decay'] = weight_decay * param_group['decay_mult']
+        i+=1
+
+    return optimizer
+
+def inv_lr_scheduler_mmd(param_lr, optimizer, iter_num, gamma, power, init_lr=0.001):
+    """Decay learning rate by a factor of 0.1 every lr_decay_epoch epochs."""
+    lr = init_lr * (1 + gamma * iter_num) ** (-power)
+
+    i=0
+    for param_group in optimizer.param_groups:
+        param_group['lr'] = lr * param_lr[i]
+        i+=1
+
+    return optimizer
+
+
+schedule_dict = {"inv": inv_lr_scheduler, "inv_mmd": inv_lr_scheduler_mmd}
--- a/network.py
+++ b/network.py
@ -0,0 +1,595 @@
+import torch.nn.functional as F
+import logging
+import numpy as np
+import torch
+import torch.nn as nn
+import torchvision
+from torchvision import models
+from torch.autograd import Variable
+import math
+import pdb
+from cvxopt import matrix, solvers
+solvers.options['show_progress'] = False
+
+
+logger = logging.getLogger(__name__)
+
+
+def calc_coeff(iter_num, high=1.0, low=0.0, alpha=10.0, max_iter=10000.0):
+    return np.float(2.0 * (high - low) / (1.0 + np.exp(-alpha*iter_num / max_iter)) - (high - low) + low)
+
+
+def init_weights(m):
+    classname = m.__class__.__name__
+    if classname.find('Conv2d') != -1 or classname.find('ConvTranspose2d') != -1:
+        nn.init.kaiming_uniform_(m.weight)
+        nn.init.zeros_(m.bias)
+    elif classname.find('BatchNorm') != -1:
+        nn.init.normal_(m.weight, 1.0, 0.02)
+        nn.init.zeros_(m.bias)
+    elif classname.find('Linear') != -1:
+        nn.init.xavier_normal_(m.weight)
+        nn.init.zeros_(m.bias)
+
+
+class RandomLayer(nn.Module):
+    def __init__(self, input_dim_list=[], output_dim=1024):
+        super(RandomLayer, self).__init__()
+        self.input_num = len(input_dim_list)
+        self.output_dim = output_dim
+        self.random_matrix = [torch.randn(input_dim_list[i], output_dim) for i in range(self.input_num)]
+
+    def forward(self, input_list):
+        return_list = [torch.mm(input_list[i], self.random_matrix[i]) for i in range(self.input_num)]
+        return_tensor = return_list[0] / math.pow(float(self.output_dim), 1.0/len(return_list))
+        for single in return_list[1:]:
+            return_tensor = torch.mul(return_tensor, single)
+        return return_tensor
+
+    def cuda(self):
+        super(RandomLayer, self).cuda()
+        self.random_matrix = [val.cuda() for val in self.random_matrix]
+
+
+class LRN(nn.Module):
+    def __init__(self, local_size=1, alpha=1.0, beta=0.75, ACROSS_CHANNELS=True):
+        super(LRN, self).__init__()
+        self.ACROSS_CHANNELS = ACROSS_CHANNELS
+        if ACROSS_CHANNELS:
+            self.average=nn.AvgPool3d(kernel_size=(local_size, 1, 1),
+                    stride=1,
+                    padding=(int((local_size-1.0)/2), 0, 0))
+        else:
+            self.average=nn.AvgPool2d(kernel_size=local_size,
+                    stride=1,
+                    padding=int((local_size-1.0)/2))
+        self.alpha = alpha
+        self.beta = beta
+
+    def forward(self, x):
+        if self.ACROSS_CHANNELS:
+            div = x.pow(2).unsqueeze(1)
+            div = self.average(div).squeeze(1)
+            div = div.mul(self.alpha).add(1.0).pow(self.beta)
+        else:
+            div = x.pow(2)
+            div = self.average(div)
+            div = div.mul(self.alpha).add(1.0).pow(self.beta)
+        x = x.div(div)
+        return x
+
+
+class AlexNet(nn.Module):
+
+    def __init__(self, num_classes=1000):
+        super(AlexNet, self).__init__()
+        self.features = nn.Sequential(
+            nn.Conv2d(3, 96, kernel_size=11, stride=4, padding=0),
+            nn.ReLU(inplace=True),
+            LRN(local_size=5, alpha=0.0001, beta=0.75),
+            nn.MaxPool2d(kernel_size=3, stride=2),
+            nn.Conv2d(96, 256, kernel_size=5, padding=2, groups=2),
+            nn.ReLU(inplace=True),
+            LRN(local_size=5, alpha=0.0001, beta=0.75),
+            nn.MaxPool2d(kernel_size=3, stride=2),
+            nn.Conv2d(256, 384, kernel_size=3, padding=1),
+            nn.ReLU(inplace=True),
+            nn.Conv2d(384, 384, kernel_size=3, padding=1, groups=2),
+            nn.ReLU(inplace=True),
+            nn.Conv2d(384, 256, kernel_size=3, padding=1, groups=2),
+            nn.ReLU(inplace=True),
+            nn.MaxPool2d(kernel_size=3, stride=2),
+        )
+        self.classifier = nn.Sequential(
+            nn.Linear(256 * 6 * 6, 4096),
+            nn.ReLU(inplace=True),
+            nn.Dropout(),
+            nn.Linear(4096, 4096),
+            nn.ReLU(inplace=True),
+            nn.Dropout(),
+            nn.Linear(4096, num_classes),
+        )
+
+    def forward(self, x):
+        x = self.features(x)
+        x = x.view(x.size(0), 256 * 6 * 6)
+        x = self.classifier(x)
+        return x
+
+
+def alexnet(pretrained=False, **kwargs):
+    r"""AlexNet model architecture from the
+    `"One weird trick..." <https://arxiv.org/abs/1404.5997>`_ paper.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = AlexNet(**kwargs)
+    if pretrained:
+        model_path = './alexnet.pth.tar'
+        pretrained_model = torch.load(model_path)
+        model.load_state_dict(pretrained_model['state_dict'])
+    return model
+
+
+# convnet without the last layer
+class AlexNetFc(nn.Module):
+  def __init__(self, use_bottleneck=True, bottleneck_dim=256, new_cls=False, class_num=1000):
+    super(AlexNetFc, self).__init__()
+    model_alexnet = alexnet(pretrained=True)
+    self.features = model_alexnet.features
+    self.classifier = nn.Sequential()
+    for i in range(6):
+      self.classifier.add_module("classifier"+str(i), model_alexnet.classifier[i])
+    self.feature_layers = nn.Sequential(self.features, self.classifier)
+
+    self.use_bottleneck = use_bottleneck
+    self.new_cls = new_cls
+    if new_cls:
+        if self.use_bottleneck:
+            self.bottleneck = nn.Linear(4096, bottleneck_dim)
+            self.fc = nn.Linear(bottleneck_dim, class_num)
+            self.bottleneck.apply(init_weights)
+            self.fc.apply(init_weights)
+            self.__in_features = bottleneck_dim
+        else:
+            self.fc = nn.Linear(4096, class_num)
+            self.fc.apply(init_weights)
+            self.__in_features = 4096
+    else:
+        self.fc = model_alexnet.classifier[6]
+        self.__in_features = 4096
+
+  def forward(self, x):
+    x = self.features(x)
+    x = x.view(x.size(0), -1)
+    x = self.classifier(x)
+    if self.use_bottleneck and self.new_cls:
+        x = self.bottleneck(x)
+    y = self.fc(x)
+    return x, y
+
+  def output_num(self):
+    return self.__in_features
+
+  def get_parameters(self):
+    if self.new_cls:
+        if self.use_bottleneck:
+            parameter_list = [{"params": self.features.parameters(), "lr_mult": 1, 'decay_mult': 2},
+                              {"params": self.classifier.parameters(), "lr_mult": 1, 'decay_mult': 2},
+                              {"params": self.bottleneck.parameters(), "lr_mult": 10, 'decay_mult': 2},
+                              {"params": self.fc.parameters(), "lr_mult": 10, 'decay_mult': 2}]
+        else:
+            parameter_list = [{"params": self.feature_layers.parameters(), "lr_mult": 1, 'decay_mult': 2},
+                              {"params": self.classifier.parameters(), "lr_mult": 1, 'decay_mult': 2},
+                              {"params": self.fc.parameters(), "lr_mult": 10, 'decay_mult': 2}]
+    else:
+        parameter_list = [
+            {"params": self.parameters(), "lr_mult": 1, 'decay_mult': 2}]
+    return parameter_list
+
+
+resnet_dict = {"ResNet18": models.resnet18, "ResNet34": models.resnet34,
+               "ResNet50": models.resnet50, "ResNet101": models.resnet101, "ResNet152": models.resnet152}
+
+
+def grl_hook(coeff):
+    def fun1(grad):
+        return -coeff*grad.clone()
+    return fun1
+
+
+class ResNetFc(nn.Module):
+    def __init__(self, resnet_name, use_bottleneck=True, bottleneck_dim=256, new_cls=False, class_num=1000, ma=0.0):
+        super(ResNetFc, self).__init__()
+        model_resnet = resnet_dict[resnet_name](pretrained=True)
+        self.conv1 = model_resnet.conv1
+        self.bn1 = model_resnet.bn1
+        self.relu = model_resnet.relu
+        self.maxpool = model_resnet.maxpool
+        self.layer1 = model_resnet.layer1
+        self.layer2 = model_resnet.layer2
+        self.layer3 = model_resnet.layer3
+        self.layer4 = model_resnet.layer4
+        self.avgpool = model_resnet.avgpool
+        self.feature_layers = nn.Sequential(self.conv1, self.bn1, self.relu, self.maxpool,
+                                            self.layer1, self.layer2, self.layer3, self.layer4, self.avgpool)
+
+        self.use_bottleneck = use_bottleneck
+        self.new_cls = new_cls
+        if new_cls:
+            if self.use_bottleneck:
+                self.bottleneck = nn.Linear(
+                    model_resnet.fc.in_features, bottleneck_dim)
+                self.fc = nn.Linear(bottleneck_dim, class_num)
+                self.bottleneck.apply(init_weights)
+                self.fc.apply(init_weights)
+                self.__in_features = bottleneck_dim
+            else:
+                self.fc = nn.Linear(model_resnet.fc.in_features, class_num)
+                self.fc.apply(init_weights)
+                self.__in_features = model_resnet.fc.in_features
+        else:
+            self.fc = model_resnet.fc
+            self.__in_features = model_resnet.fc.in_features
+
+        self.im_weights_update = create_im_weights_update(self, ma, class_num)
+
+    def forward(self, x):
+        x = self.feature_layers(x)
+        x = x.view(x.size(0), -1)
+        if self.use_bottleneck and self.new_cls:
+            x = self.bottleneck(x)
+        y = self.fc(x)
+        return x, y
+
+    def output_num(self):
+        return self.__in_features
+
+    def get_parameters(self):
+        if self.new_cls:
+            if self.use_bottleneck:
+                parameter_list = [{"params": self.feature_layers.parameters(), "lr_mult": 1, 'decay_mult': 2},
+                                  {"params": self.bottleneck.parameters(
+                                  ), "lr_mult": 10, 'decay_mult': 2},
+                                  {"params": self.fc.parameters(), "lr_mult": 10,
+                                   'decay_mult': 2},
+                                  {"params": self.im_weights, "lr_mult": 10, 'decay_mult': 2}]
+            else:
+                parameter_list = [{"params": self.feature_layers.parameters(), "lr_mult": 1, 'decay_mult': 2},
+                                  {"params": self.fc.parameters(), "lr_mult": 10,
+                                   'decay_mult': 2},
+                                  {"params": self.im_weights, "lr_mult": 10, 'decay_mult': 2}]
+        else:
+            parameter_list = [{"params": self.parameters(), "lr_mult": 1, 'decay_mult': 2},
+                              {"params": self.im_weights, "lr_mult": 10, 'decay_mult': 2}]
+        return parameter_list
+
+
+vgg_dict = {"VGG11": models.vgg11, "VGG13": models.vgg13, "VGG16": models.vgg16, "VGG19": models.vgg19,
+            "VGG11BN": models.vgg11_bn, "VGG13BN": models.vgg13_bn, "VGG16BN": models.vgg16_bn, "VGG19BN": models.vgg19_bn}
+
+
+class VGGFc(nn.Module):
+  def __init__(self, vgg_name, use_bottleneck=True, bottleneck_dim=256, new_cls=False, class_num=1000):
+    super(VGGFc, self).__init__()
+    model_vgg = vgg_dict[vgg_name](pretrained=True)
+    self.features = model_vgg.features
+    self.classifier = nn.Sequential()
+    for i in range(6):
+        self.classifier.add_module(
+            "classifier"+str(i), model_vgg.classifier[i])
+    self.feature_layers = nn.Sequential(self.features, self.classifier)
+
+    self.use_bottleneck = use_bottleneck
+    self.new_cls = new_cls
+    if new_cls:
+        if self.use_bottleneck:
+            self.bottleneck = nn.Linear(4096, bottleneck_dim)
+            self.fc = nn.Linear(bottleneck_dim, class_num)
+            self.bottleneck.apply(init_weights)
+            self.fc.apply(init_weights)
+            self.__in_features = bottleneck_dim
+        else:
+            self.fc = nn.Linear(4096, class_num)
+            self.fc.apply(init_weights)
+            self.__in_features = 4096
+    else:
+        self.fc = model_vgg.classifier[6]
+        self.__in_features = 4096
+
+  def forward(self, x):
+    x = self.features(x)
+    x = x.view(x.size(0), -1)
+    x = self.classifier(x)
+    if self.use_bottleneck and self.new_cls:
+        x = self.bottleneck(x)
+    y = self.fc(x)
+    return x, y
+
+  def output_num(self):
+    return self.__in_features
+
+  def get_parameters(self):
+    if self.new_cls:
+        if self.use_bottleneck:
+            parameter_list = [{"params": self.features.parameters(), "lr_mult": 1, 'decay_mult': 2},
+                              {"params": self.classifier.parameters(
+                              ), "lr_mult": 1, 'decay_mult': 2},
+                              {"params": self.bottleneck.parameters(
+                              ), "lr_mult": 10, 'decay_mult': 2},
+                              {"params": self.fc.parameters(), "lr_mult": 10, 'decay_mult': 2}]
+        else:
+            parameter_list = [{"params": self.feature_layers.parameters(), "lr_mult": 1, 'decay_mult': 2},
+                              {"params": self.classifier.parameters(
+                              ), "lr_mult": 1, 'decay_mult': 2},
+                              {"params": self.fc.parameters(), "lr_mult": 10, 'decay_mult': 2}]
+    else:
+        parameter_list = [
+            {"params": self.parameters(), "lr_mult": 1, 'decay_mult': 2}]
+    return parameter_list
+
+# For SVHN dataset
+
+
+class DTN(nn.Module):
+    def __init__(self, ma=0.0):
+        super(DTN, self).__init__()
+        self.conv_params = nn.Sequential(
+            nn.Conv2d(3, 64, kernel_size=5, stride=2, padding=2),
+            nn.BatchNorm2d(64),
+            nn.Dropout2d(0.1),
+            nn.ReLU(),
+            nn.Conv2d(64, 128, kernel_size=5, stride=2, padding=2),
+            nn.BatchNorm2d(128),
+            nn.Dropout2d(0.3),
+            nn.ReLU(),
+            nn.Conv2d(128, 256, kernel_size=5, stride=2, padding=2),
+            nn.BatchNorm2d(256),
+            nn.Dropout2d(0.5),
+            nn.ReLU()
+        )
+
+        self.fc_params = nn.Sequential(
+            nn.Linear(256*4*4, 512),
+            nn.BatchNorm1d(512),
+            nn.ReLU(),
+            nn.Dropout()
+        )
+
+        class_num = 10
+
+        self.classifier = nn.Linear(512, class_num)
+        self.__in_features = 512
+
+        self.im_weights_update = create_im_weights_update(self, ma, class_num)
+
+    def forward(self, x):
+        x = self.conv_params(x)
+        x = x.view(x.size(0), -1)
+        x = self.fc_params(x)
+        y = self.classifier(x)
+        return x, y
+
+    def output_num(self):
+        return self.__in_features
+
+
+class LeNet(nn.Module):
+    def __init__(self, ma=0.0):
+        super(LeNet, self).__init__()
+        self.conv_params = nn.Sequential(
+                nn.Conv2d(1, 20, kernel_size=5),
+                nn.MaxPool2d(2),
+                nn.ReLU(),
+                nn.Conv2d(20, 50, kernel_size=5),
+                nn.Dropout2d(p=0.5),
+                nn.MaxPool2d(2),
+                nn.ReLU(),
+                )
+
+        class_num = 10
+
+        self.fc_params = nn.Sequential(nn.Linear(50*4*4, 500), nn.ReLU(), nn.Dropout(p=0.5))
+        self.classifier = nn.Linear(500, class_num)
+        self.__in_features = 500
+
+        self.im_weights_update = create_im_weights_update(self, ma, class_num)
+
+    def forward(self, x):
+        x = self.conv_params(x)
+        x = x.view(x.size(0), -1)
+        x = self.fc_params(x)
+        y = self.classifier(x)
+        return x, y
+
+    def output_num(self):
+        return self.__in_features
+
+
+class AdversarialNetwork(nn.Module):
+  def __init__(self, in_feature, hidden_size, sigmoid=True):
+    super(AdversarialNetwork, self).__init__()
+    self.ad_layer1 = nn.Linear(in_feature, hidden_size)
+    self.ad_layer2 = nn.Linear(hidden_size, hidden_size)
+    self.ad_layer3 = nn.Linear(hidden_size, 1)
+    self.relu1 = nn.ReLU()
+    self.relu2 = nn.ReLU()
+    self.dropout1 = nn.Dropout(0.5)
+    self.dropout2 = nn.Dropout(0.5)
+    self.sigmoid = sigmoid
+    self.apply(init_weights)
+    self.iter_num = 0
+    self.alpha = 10
+    self.low = 0.0
+    self.high = 1.0
+    self.max_iter = 10000.0
+
+  def forward(self, x):
+    if self.training:
+        self.iter_num += 1
+    coeff = calc_coeff(self.iter_num, self.high, self.low, self.alpha, self.max_iter)
+    x = x * 1.0
+    x.register_hook(grl_hook(coeff))
+    x = self.ad_layer1(x)
+    x = self.relu1(x)
+    x = self.dropout1(x)
+    x = self.ad_layer2(x)
+    x = self.relu2(x)
+    x = self.dropout2(x)
+    y = self.ad_layer3(x)
+    if self.sigmoid:
+        y = nn.Sigmoid()(y)
+    return y
+
+  def output_num(self):
+    return 1
+
+  def get_parameters(self):
+    return [{"params": self.parameters(), "lr_mult": 10, 'decay_mult': 2}]
+
+
+class GradientReversalLayer(torch.autograd.Function):
+    """
+    Implement the gradient reversal layer for the convenience of domain adaptation neural network.
+    The forward part is the identity function while the backward part is the negative function.
+    """
+    @staticmethod
+    def forward(ctx, inputs):
+        return inputs.view_as(inputs)
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        return grad_output.neg()
+
+
+def grad_reverse(tensor):
+    return GradientReversalLayer.apply(tensor)
+
+
+class ConvNet(nn.Module):
+    """
+    Vanilla CNN for classification.
+    """
+
+    def __init__(self, configs):
+        super(ConvNet, self).__init__()
+        self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
+        self.conv2 = nn.Conv2d(10, 20, kernel_size=5)
+        self.fc = nn.Linear(1280, 100)
+        self.softmax = nn.Linear(100, configs["num_classes"])
+        self.num_classes = configs["num_classes"]
+
+    def forward(self, inputs):
+        feats = F.relu(self.conv1(inputs))
+        feats = F.relu(self.conv2(feats))
+        feats = feats.view(-1, 1280)
+        feats = F.relu(self.fc(feats))
+        logprobs = F.log_softmax(self.softmax(feats), dim=1)
+        return logprobs
+
+
+class ResNet50Fc(nn.Module):
+  def __init__(self, ma=0.0, class_num=31, **kwargs):
+    super(ResNet50Fc, self).__init__()
+    model_resnet50 = models.resnet50(pretrained=True)
+    self.conv1 = model_resnet50.conv1
+    self.bn1 = model_resnet50.bn1
+    self.relu = model_resnet50.relu
+    self.maxpool = model_resnet50.maxpool
+    self.layer1 = model_resnet50.layer1
+    self.layer2 = model_resnet50.layer2
+    self.layer3 = model_resnet50.layer3
+    self.layer4 = model_resnet50.layer4
+    self.avgpool = model_resnet50.avgpool
+    self.__in_features = model_resnet50.fc.in_features
+
+    self.im_weights_update = create_im_weights_update(self, ma, class_num)
+
+  def forward(self, x):
+    x = self.conv1(x)
+    x = self.bn1(x)
+    x = self.relu(x)
+    x = self.maxpool(x)
+    x = self.layer1(x)
+    x = self.layer2(x)
+    x = self.layer3(x)
+    x = self.layer4(x)
+    x = self.avgpool(x)
+    x = x.view(x.size(0), -1)
+    return x
+
+  def output_num(self):
+    return self.__in_features
+
+
+class LeNetMMD(nn.Module):
+    def __init__(self, ma=0.0, **kwargs):
+        super(LeNetMMD, self).__init__()
+        self.conv_params = nn.Sequential(
+            nn.Conv2d(1, 20, kernel_size=5),
+            nn.MaxPool2d(2),
+            nn.ReLU(),
+            nn.Conv2d(20, 50, kernel_size=5),
+            nn.Dropout2d(p=0.5),
+            nn.MaxPool2d(2),
+            nn.ReLU(),
+        )
+
+        class_num = 10
+
+        self.fc_params = nn.Sequential(
+            nn.Linear(50*4*4, 500), nn.ReLU(), nn.Dropout(p=0.5))
+        # self.__in_features = 500
+        self.__in_features = 800
+
+        self.im_weights_update = create_im_weights_update(self, ma, class_num)
+
+    def forward(self, x):
+        x = self.conv_params(x)
+        x = x.view(x.size(0), -1)
+        # x = self.fc_params(x)
+        return x
+
+    def output_num(self):
+        return self.__in_features
+
+
+network_dict = {"ResNet50" : ResNet50Fc, "LeNet": LeNetMMD}
+
+
+def create_im_weights_update(class_inst, ma, class_num):
+
+    # Label importance weights.
+    class_inst.ma = ma
+    class_inst.im_weights = nn.Parameter(
+        torch.ones(class_num, 1), requires_grad=False)
+
+    def im_weights_update(source_y, target_y, cov, device, inst=class_inst):
+        """
+        Solve a Quadratic Program to compute the optimal importance weight under the generalized label shift assumption.
+        :param source_y:    The marginal label distribution of the source domain.
+        :param target_y:    The marginal pseudo-label distribution of the target domain from the current classifier.
+        :param cov:         The covariance matrix of predicted-label and true label of the source domain.
+        :param device:      Device of the operation.
+        :return:
+        """
+        # Convert all the vectors to column vectors.
+        dim = cov.shape[0]
+        source_y = source_y.reshape(-1, 1).astype(np.double)
+        target_y = target_y.reshape(-1, 1).astype(np.double)
+        cov = cov.astype(np.double)
+
+        P = matrix(np.dot(cov.T, cov), tc="d")
+        q = -matrix(np.dot(cov, target_y), tc="d")
+        G = matrix(-np.eye(dim), tc="d")
+        h = matrix(np.zeros(dim), tc="d")
+        A = matrix(source_y.reshape(1, -1), tc="d")
+        b = matrix([1.0], tc="d")
+        sol = solvers.qp(P, q, G, h, A, b)
+        new_im_weights = np.array(sol["x"])
+
+        # EMA for the weights
+        inst.im_weights.data = (1 - inst.ma) * torch.tensor(
+            new_im_weights, dtype=torch.float32).to(device) + inst.ma * inst.im_weights.data
+
+    return im_weights_update
--- a/pre_process.py
+++ b/pre_process.py
@ -0,0 +1,256 @@
+import numpy as np
+from torchvision import transforms
+import os
+from PIL import Image, ImageOps
+import numbers
+import torch
+
+class ResizeImage():
+    def __init__(self, size):
+      if isinstance(size, int):
+        self.size = (int(size), int(size))
+      else:
+        self.size = size
+    def __call__(self, img):
+      th, tw = self.size
+      return img.resize((th, tw))
+
+class RandomSizedCrop(object):
+    """Crop the given PIL.Image to random size and aspect ratio.
+    A crop of random size of (0.08 to 1.0) of the original size and a random
+    aspect ratio of 3/4 to 4/3 of the original aspect ratio is made. This crop
+    is finally resized to given size.
+    This is popularly used to train the Inception networks.
+    Args:
+        size: size of the smaller edge
+        interpolation: Default: PIL.Image.BILINEAR
+    """
+
+    def __init__(self, size, interpolation=Image.BILINEAR):
+        self.size = size
+        self.interpolation = interpolation
+
+    def __call__(self, img):
+        h_off = random.randint(0, img.shape[1]-self.size)
+        w_off = random.randint(0, img.shape[2]-self.size)
+        img = img[:, h_off:h_off+self.size, w_off:w_off+self.size]
+        return img
+
+
+class Normalize(object):
+    """Normalize an tensor image with mean and standard deviation.
+    Given mean: (R, G, B),
+    will normalize each channel of the torch.*Tensor, i.e.
+    channel = channel - mean
+    Args:
+        mean (sequence): Sequence of means for R, G, B channels respecitvely.
+    """
+
+    def __init__(self, mean=None, meanfile=None):
+        if mean:
+            self.mean = mean
+        else:
+            arr = np.load(meanfile)
+            self.mean = torch.from_numpy(arr.astype('float32')/255.0)[[2,1,0],:,:]
+
+    def __call__(self, tensor):
+        """
+        Args:
+            tensor (Tensor): Tensor image of size (C, H, W) to be normalized.
+        Returns:
+            Tensor: Normalized image.
+        """
+        # TODO: make efficient
+        for t, m in zip(tensor, self.mean):
+            t.sub_(m)
+        return tensor
+
+
+
+class PlaceCrop(object):
+    """Crops the given PIL.Image at the particular index.
+    Args:
+        size (sequence or int): Desired output size of the crop. If size is an
+            int instead of sequence like (w, h), a square crop (size, size) is
+            made.
+    """
+
+    def __init__(self, size, start_x, start_y):
+        if isinstance(size, int):
+            self.size = (int(size), int(size))
+        else:
+            self.size = size
+        self.start_x = start_x
+        self.start_y = start_y
+
+    def __call__(self, img):
+        """
+        Args:
+            img (PIL.Image): Image to be cropped.
+        Returns:
+            PIL.Image: Cropped image.
+        """
+        th, tw = self.size
+        return img.crop((self.start_x, self.start_y, self.start_x + tw, self.start_y + th))
+
+
+class ForceFlip(object):
+    """Horizontally flip the given PIL.Image randomly with a probability of 0.5."""
+
+    def __call__(self, img):
+        """
+        Args:
+            img (PIL.Image): Image to be flipped.
+        Returns:
+            PIL.Image: Randomly flipped image.
+        """
+        return img.transpose(Image.FLIP_LEFT_RIGHT)
+
+
+class CenterCrop(object):
+    """Crops the given PIL.Image at the center.
+    Args:
+        size (sequence or int): Desired output size of the crop. If size is an
+            int instead of sequence like (h, w), a square crop (size, size) is
+            made.
+    """
+
+    def __init__(self, size):
+        if isinstance(size, numbers.Number):
+            self.size = (int(size), int(size))
+        else:
+            self.size = size
+
+    def __call__(self, img):
+        """
+        Args:
+            img (PIL.Image): Image to be cropped.
+        Returns:
+            PIL.Image: Cropped image.
+        """
+        w, h = (img.shape[1], img.shape[2])
+        th, tw = self.size
+        w_off = int((w - tw) / 2.)
+        h_off = int((h - th) / 2.)
+        img = img[:, h_off:h_off+th, w_off:w_off+tw]
+        return img
+
+
+def image_train(resize_size=256, crop_size=224, alexnet=False, LeNet=False):
+
+  if LeNet:
+      return transforms.Compose([
+          transforms.Resize((28, 28)),
+          transforms.ToTensor(),
+          transforms.Normalize((0.5,), (0.5,))])
+
+  if not alexnet:
+    normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
+                                   std=[0.229, 0.224, 0.225])
+  else:
+    normalize = Normalize(meanfile='./ilsvrc_2012_mean.npy')
+  return  transforms.Compose([
+        ResizeImage(resize_size),
+        transforms.RandomResizedCrop(crop_size),
+        transforms.RandomHorizontalFlip(),
+        transforms.ToTensor(),
+        normalize
+    ])
+
+
+def image_test(resize_size=256, crop_size=224, alexnet=False, LeNet=False):
+
+  if LeNet:
+      return transforms.Compose([
+          transforms.Resize((28, 28)),
+          transforms.ToTensor(),
+          transforms.Normalize((0.5,), (0.5,))])
+
+  if not alexnet:
+    normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
+                                   std=[0.229, 0.224, 0.225])
+  else:
+    normalize = Normalize(meanfile='./ilsvrc_2012_mean.npy')
+  start_first = 0
+  start_center = (resize_size - crop_size - 1) / 2
+  start_last = resize_size - crop_size - 1
+
+  return transforms.Compose([
+    ResizeImage(resize_size),
+    PlaceCrop(crop_size, start_center, start_center),
+    transforms.ToTensor(),
+    normalize
+  ])
+
+def image_test_10crop(resize_size=256, crop_size=224, alexnet=False):
+    if not alexnet:
+        normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
+                                   std=[0.229, 0.224, 0.225])
+    else:
+        normalize = Normalize(meanfile='./ilsvrc_2012_mean.npy')
+    start_first = 0
+    start_center = (resize_size - crop_size - 1) / 2
+    start_last = resize_size - crop_size - 1
+    data_transforms = [
+        transforms.Compose([
+        ResizeImage(resize_size),ForceFlip(),
+        PlaceCrop(crop_size, start_first, start_first),
+        transforms.ToTensor(),
+        normalize
+        ]),
+        transforms.Compose([
+        ResizeImage(resize_size),ForceFlip(),
+        PlaceCrop(crop_size, start_last, start_last),
+        transforms.ToTensor(),
+        normalize
+        ]),
+        transforms.Compose([
+        ResizeImage(resize_size),ForceFlip(),
+        PlaceCrop(crop_size, start_last, start_first),
+        transforms.ToTensor(),
+        normalize
+        ]),
+        transforms.Compose([
+        ResizeImage(resize_size),ForceFlip(),
+        PlaceCrop(crop_size, start_first, start_last),
+        transforms.ToTensor(),
+        normalize
+        ]),
+        transforms.Compose([
+        ResizeImage(resize_size),ForceFlip(),
+        PlaceCrop(crop_size, start_center, start_center),
+        transforms.ToTensor(),
+        normalize
+        ]),
+        transforms.Compose([
+        ResizeImage(resize_size),
+        PlaceCrop(crop_size, start_first, start_first),
+        transforms.ToTensor(),
+        normalize
+        ]),
+        transforms.Compose([
+        ResizeImage(resize_size),
+        PlaceCrop(crop_size, start_last, start_last),
+        transforms.ToTensor(),
+        normalize
+        ]),
+        transforms.Compose([
+        ResizeImage(resize_size),
+        PlaceCrop(crop_size, start_last, start_first),
+        transforms.ToTensor(),
+        normalize
+        ]),
+        transforms.Compose([
+        ResizeImage(resize_size),
+        PlaceCrop(crop_size, start_first, start_last),
+        transforms.ToTensor(),
+        normalize
+        ]),
+        transforms.Compose([
+        ResizeImage(resize_size),
+        PlaceCrop(crop_size, start_center, start_center),
+        transforms.ToTensor(),
+        normalize
+        ])
+    ]
+    return data_transforms
--- a/train_digits.py
+++ b/train_digits.py
@ -0,0 +1,391 @@
+import argparse
+import numpy as np
+import os
+import pickle
+import scipy.stats
+import sys
+import time
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.optim as optim
+from torchvision import datasets, transforms
+
+import loss as loss_func
+import network
+from data_list import build_uspsmnist, sample_ratios, subsampling
+
+
+def write_list(f, l):
+    f.write(",".join(map(str, l)) + "\n")
+    f.flush()
+    sys.stdout.flush()
+
+
+def train(args, model, ad_net,
+          source_samples, source_labels, target_samples, target_labels, optimizer, optimizer_ad,
+          epoch, start_epoch, method,
+          source_label_distribution, out_wei_file,
+          cov_mat, pseudo_target_label, class_weights, true_weights):
+    model.train()
+
+    cov_mat[:] = 0.0
+    pseudo_target_label[:] = 0.0
+
+    len_source = source_labels.shape[0]
+    len_target = target_labels.shape[0]
+
+    size = max(len_source, len_target)
+    num_iter = int(size / args.batch_size)
+
+    for batch_idx in range(num_iter):
+        t = time.time()
+        source_idx = np.random.choice(len_source, args.batch_size)
+        target_idx = np.random.choice(len_target, args.batch_size)
+        data_source, label_source = source_samples[source_idx], source_labels[source_idx]
+        data_target, _ = target_samples[target_idx], target_labels[target_idx]
+
+        optimizer.zero_grad()
+        optimizer_ad.zero_grad()
+        feature, output = model(torch.cat((data_source, data_target), 0))
+
+        if 'IW' in method:
+            ys_onehot = torch.zeros(args.batch_size, 10).to(args.device)
+            ys_onehot.scatter_(1, label_source.view(-1, 1), 1)
+            # Compute weights on source data.
+            if 'ORACLE' in method:
+                weights = torch.mm(ys_onehot, true_weights)
+            else:
+                weights = torch.mm(ys_onehot, model.im_weights)
+
+            source_preds, target_preds = output[:
+                                                args.batch_size], output[args.batch_size:]
+            # Compute the aggregated distribution of pseudo-label on the target domain.
+            pseudo_target_label += torch.sum(
+                F.softmax(target_preds, dim=1), dim=0).view(-1, 1).detach()
+            # Update the covariance matrix on the source domain as well.
+            cov_mat += torch.mm(F.softmax(source_preds,
+                                            dim=1).transpose(1, 0), ys_onehot).detach()
+
+            loss = torch.mean(
+                nn.CrossEntropyLoss(weight=class_weights, reduction='none')
+                (output.narrow(0, 0, data_source.size(0)), label_source) * weights) / 10.0
+        else:
+            loss = nn.CrossEntropyLoss()(output.narrow(0, 0, data_source.size(0)), label_source)
+
+        if epoch > start_epoch:
+            if method == 'CDAN-E':
+                softmax_output = nn.Softmax(dim=1)(output)
+                entropy = loss_func.Entropy(softmax_output)
+                loss += loss_func.CDAN([feature, softmax_output], ad_net, entropy, network.calc_coeff(
+                    num_iter*(epoch-start_epoch)+batch_idx), None, device=args.device)
+
+            elif 'IWCDAN-E' in method:
+                    softmax_output = nn.Softmax(dim=1)(output)
+                    entropy = loss_func.Entropy(softmax_output)
+                    loss += loss_func.CDAN([feature, softmax_output], ad_net, entropy, network.calc_coeff(
+                        num_iter*(epoch-start_epoch)+batch_idx), None, weights=weights, device=args.device)
+
+            elif method == 'CDAN':
+                    softmax_output = nn.Softmax(dim=1)(output)
+                    loss += loss_func.CDAN([feature, softmax_output],
+                                        ad_net, None, None, None, device=args.device)
+
+            elif 'IWCDAN' in method:
+                    softmax_output = nn.Softmax(dim=1)(output)
+                    loss += loss_func.CDAN([feature, softmax_output],
+                                        ad_net, None, None, None, weights=weights, device=args.device)
+
+            elif method == 'DANN':
+                    loss += loss_func.DANN(feature, ad_net, args.device)
+
+            elif 'IWDAN' in method:
+                    dloss = loss_func.IWDAN(feature, ad_net, weights)
+                    loss += args.mu * dloss
+
+            elif method == 'NANN':
+                pass
+
+            else:
+                raise ValueError('Method cannot be recognized.')
+
+        loss.backward()
+        optimizer.step()
+
+        if epoch > start_epoch and method != 'NANN':
+            optimizer_ad.step()
+
+    if 'IW' in method  and epoch > start_epoch:
+        pseudo_target_label /= args.batch_size * num_iter
+        cov_mat /= args.batch_size * num_iter
+        # Recompute the importance weight by solving a QP.
+        model.im_weights_update(source_label_distribution,
+                                pseudo_target_label.cpu().detach().numpy(),
+                                cov_mat.cpu().detach().numpy(),
+                                args.device)
+        current_weights = [round(x, 4) for x in model.im_weights.data.cpu().numpy().flatten()]
+        write_list(out_wei_file, [np.linalg.norm(
+            current_weights - true_weights.cpu().numpy().flatten())] + current_weights)
+        print(np.linalg.norm(current_weights - true_weights.cpu().numpy().flatten()), current_weights)
+
+
+def test(args, epoch, model, test_samples, test_labels, start_time_test, out_log_file, name=''):
+    model.eval()
+    test_loss = 0
+    correct = 0
+    len_test = test_labels.shape[0]
+
+    for i in range(len_test):
+        data, target = test_samples[i].unsqueeze(0), test_labels[i].unsqueeze(0)
+        _, output = model(data)
+        test_loss += nn.CrossEntropyLoss()(output, target).item()
+        pred = output.data.cpu().max(1, keepdim=True)[1]
+        correct += pred.eq(target.data.cpu().view_as(pred)).sum().item()
+
+    test_loss /= len_test
+    temp_acc = 100. * correct / len_test
+    log_str = "  {}, iter: {:05d}, sec: {:.0f}, loss: {:.5f}, accuracy: {}/{}, precision: {:.5f}".format(name, epoch, time.time() - start_time_test, test_loss, correct, len_test, temp_acc)
+    print(log_str)
+    sys.stdout.flush()
+    out_log_file.write(log_str+"\n")
+    out_log_file.flush()
+
+
+def main():
+    # Training settings
+    parser = argparse.ArgumentParser(description='CDAN USPS MNIST')
+    parser.add_argument('method', type=str, default='CDAN-E',
+                        choices=['CDAN', 'CDAN-E', 'DANN', 'IWDAN', 'NANN', 'IWDANORACLE', 'IWCDAN', 'IWCDANORACLE', 'IWCDAN-E', 'IWCDAN-EORACLE'])
+    parser.add_argument('--task', default='mnist2usps', help='task to perform', choices=['usps2mnist', 'mnist2usps'])
+    parser.add_argument('--batch_size', type=int, default=64,
+                        help='input batch size for training (default: 64)')
+    parser.add_argument('--test_batch_size', type=int, default=1000,
+                        help='input batch size for testing (default: 1000)')
+    parser.add_argument('--epochs', type=int, default=70, metavar='N',
+                        help='number of epochs to train (default: 70)')
+    parser.add_argument('--lr', type=float, default=0.0, metavar='LR',
+                        help='learning rate (default: 0.02)')
+    parser.add_argument('--momentum', type=float, default=0.5, metavar='M',
+                        help='SGD momentum (default: 0.5)')
+    parser.add_argument('--seed', type=int, default=42, metavar='S',
+                        help='random seed (default: 42)')
+    parser.add_argument('--log_interval', type=int, default=50,
+                        help='how many batches to wait before logging training status')
+    parser.add_argument('--root_folder', type=str, default='data/usps2mnist/', help="The folder containing the datasets and the lists")
+    parser.add_argument('--output_dir', type=str, default='results', help="output directory")
+    parser.add_argument("-u", "--mu", help="Hyperparameter of the coefficient of the domain adversarial loss", type=float, default=1.0)
+    parser.add_argument('--ratio', type =float, default=0, help='ratio option')
+    parser.add_argument('--ma', type=float, default=0.5, help='weight for the moving average of iw')
+    args = parser.parse_args()
+    args.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+    # Running the JSD experiment on fewer epochs for efficiency
+    if args.ratio >= 100:
+        args.epochs = 25
+
+    print('Running {} on {} for {} epochs on task {}'.format(
+        args.method, args.device, args.epochs, args.task))
+
+    # Set random number seed.
+    np.random.seed(args.seed)
+    torch.manual_seed(args.seed)
+    os.environ["CUDA_VISIBLE_DEVICES"] = '0'
+
+    if args.task == 'usps2mnist':
+
+        # CDAN parameters
+        decay_epoch = 6
+        decay_frac = 0.5
+        lr = 0.02
+        start_epoch = 1
+        model = network.LeNet(args.ma)
+        build_dataset = build_uspsmnist
+
+        source_list = os.path.join(args.root_folder, 'usps_train.txt')
+        source_path = os.path.join(args.root_folder, 'usps_train_dataset.pkl')
+        target_list = os.path.join(args.root_folder, 'mnist_train.txt')
+        target_path = os.path.join(args.root_folder, 'mnist_train_dataset.pkl')
+        test_list   = os.path.join(args.root_folder, 'mnist_test.txt')
+        test_path   = os.path.join(args.root_folder, 'mnist_test_dataset.pkl')
+
+    elif args.task == 'mnist2usps':
+
+        decay_epoch = 5
+        decay_frac = 0.5
+        lr = 0.02
+        start_epoch = 1
+        model = network.LeNet(args.ma)
+        build_dataset = build_uspsmnist
+
+        source_list = os.path.join(args.root_folder, 'mnist_train.txt')
+        source_path = os.path.join(args.root_folder, 'mnist_train_dataset.pkl')
+        target_list = os.path.join(args.root_folder, 'usps_train.txt')
+        target_path = os.path.join(args.root_folder, 'usps_train_dataset.pkl')
+        test_list   = os.path.join(args.root_folder, 'usps_test.txt')
+        test_path   = os.path.join(args.root_folder, 'usps_test_dataset.pkl')
+
+    else:
+        raise Exception('Task cannot be recognized!')
+
+    out_log_file = open(os.path.join(args.output_dir, "log.txt"), "w")
+    out_log_file_train = open(os.path.join(
+        args.output_dir, "log_train.txt"), "w")
+    if not os.path.exists(args.output_dir):
+        os.mkdir(args.output_dir)
+
+    model = model.to(args.device)
+    class_num = 10
+
+    if args.lr > 0:
+      lr = args.lr
+
+    print('Starting loading data')
+    sys.stdout.flush()
+    t_data = time.time()
+    if os.path.exists(source_path):
+        print('Found existing dataset for source')
+        with open(source_path, 'rb') as f:
+            [source_samples, source_labels] = pickle.load(f)
+            source_samples, source_labels = torch.Tensor(source_samples).to(
+                args.device), torch.LongTensor(source_labels).to(args.device)
+    else:
+        print('Building dataset for source and writing to {}'.format(source_path))
+        source_samples, source_labels = build_dataset(
+            source_list, source_path, args.root_folder, args.device)
+
+    if os.path.exists(target_path):
+        print('Found existing dataset for target')
+        with open(target_path, 'rb') as f:
+            [target_samples, target_labels] = pickle.load(f)
+            target_samples, target_labels = torch.Tensor(
+                target_samples).to(args.device), torch.LongTensor(target_labels).to(args.device)
+    else:
+        print('Building dataset for target and writing to {}'.format(target_path))
+        target_samples, target_labels = build_dataset(
+            target_list, target_path, args.root_folder, args.device)
+
+    if os.path.exists(test_path):
+        print('Found existing dataset for test')
+        with open(test_path, 'rb') as f:
+            [test_samples, test_labels] = pickle.load(f)
+            test_samples, test_labels = torch.Tensor(
+                test_samples).to(args.device), torch.LongTensor(test_labels).to(args.device)
+    else:
+        print('Building dataset for test and writing to {}'.format(test_path))
+        test_samples, test_labels = build_dataset(
+            test_list, test_path, args.root_folder, args.device)
+
+    print('Data loaded in {}'.format(time.time() - t_data))
+
+    if args.ratio == 1:
+        # RATIO OPTION 1
+        # 30% of the samples from the first 5 classes
+        print('Using option 1, ie [0.3] * 5 + [1] * 5')
+        ratios_source = [0.3] * 5 + [1] * 5
+        ratios_target = [1] * 10
+    elif args.ratio >= 200:
+        s_ = subsampling[int(args.ratio) % 100]
+        ratios_source = s_[0]
+        ratios_target = [1] * 10
+        print('Using random subset ratio {} of the source, with theoretical jsd {}'.format(args.ratio, s_[1]))
+    elif 200 > args.ratio >= 100:
+        s_ = subsampling[int(args.ratio) % 100]
+        ratios_source = [1] * 10
+        ratios_target = s_[0]
+        print('Using random subset ratio {} of the target, with theoretical jsd {}'.format(args.ratio, s_[1]))
+    else:
+        # ORIGINAL DATASETS
+        print('Using original datasets')
+        ratios_source = [1] * 10
+        ratios_target = [1] * 10
+    ratios_test = ratios_target
+
+    # Subsample dataset if need be
+    source_samples, source_labels = sample_ratios(
+        source_samples, source_labels, ratios_source)
+    target_samples, target_labels = sample_ratios(
+        target_samples, target_labels, ratios_target)
+    test_samples, test_labels = sample_ratios(
+        test_samples, test_labels, ratios_test)
+
+    # compute labels distribution on the source and target domain
+    source_label_distribution = np.zeros((class_num))
+    for img in source_labels:
+        source_label_distribution[int(img.item())] += 1
+    print("Total source samples: {}".format(
+        np.sum(source_label_distribution)), flush=True)
+    print("Source samples per class: {}".format(source_label_distribution))
+    source_label_distribution /= np.sum(source_label_distribution)
+    write_list(out_log_file, source_label_distribution)
+    print("Source label distribution: {}".format(source_label_distribution))
+    target_label_distribution = np.zeros((class_num))
+    for img in target_labels:
+        target_label_distribution[int(img.item())] += 1
+    print("Total target samples: {}".format(
+        np.sum(target_label_distribution)), flush=True)
+    print("Target samples per class: {}".format(target_label_distribution))
+    target_label_distribution /= np.sum(target_label_distribution)
+    write_list(out_log_file, target_label_distribution)
+    print("Target label distribution: {}".format(target_label_distribution))
+    test_label_distribution = np.zeros((class_num))
+    for img in test_labels:
+        test_label_distribution[int(img.item())] += 1
+    print("Test samples per class: {}".format(test_label_distribution))
+    test_label_distribution /= np.sum(test_label_distribution)
+    write_list(out_log_file, test_label_distribution)
+    print("Test label distribution: {}".format(test_label_distribution))
+    mixture = (source_label_distribution + target_label_distribution) / 2
+    jsd = (scipy.stats.entropy(source_label_distribution, qk=mixture)
+           + scipy.stats.entropy(target_label_distribution, qk=mixture)) / 2
+    print("JSD source to target : {}".format(jsd))
+    mixture_2 = (test_label_distribution + target_label_distribution) / 2
+    jsd_2 = (scipy.stats.entropy(test_label_distribution, qk=mixture_2)
+           + scipy.stats.entropy(target_label_distribution, qk=mixture_2)) / 2
+    print("JSD test to target : {}".format(jsd_2))
+    out_wei_file = open(os.path.join(args.output_dir, "log_weights_{}.txt".format(jsd)), "w")
+    write_list(out_wei_file, [round(x, 4) for x in source_label_distribution])
+    write_list(out_wei_file, [round(x, 4) for x in target_label_distribution])
+    out_wei_file.write(str(jsd) + "\n")
+    true_weights = torch.tensor(
+        target_label_distribution / source_label_distribution, dtype=torch.float, requires_grad=False)[:, None].to(args.device)
+    print("True weights : {}".format(true_weights[:, 0].cpu().numpy()))
+
+    if 'CDAN' in args.method:
+        ad_net = network.AdversarialNetwork(
+            model.output_num() * class_num, 500, sigmoid='WDANN' not in args.method)
+    else:
+        ad_net = network.AdversarialNetwork(
+            model.output_num(), 500, sigmoid='WDANN' not in args.method)
+
+    ad_net = ad_net.to(args.device)
+
+    optimizer = optim.SGD(model.parameters(), lr=lr,
+                          weight_decay=0.0005, momentum=0.9)
+    optimizer_ad = optim.SGD(
+        ad_net.parameters(), lr=lr, weight_decay=0.0005, momentum=0.9)
+
+    # Maintain two quantities for the QP.
+    cov_mat = torch.tensor(np.zeros((class_num, class_num), dtype=np.float32),
+                           requires_grad=False).to(args.device)
+    pseudo_target_label = torch.tensor(np.zeros((class_num, 1), dtype=np.float32),
+                                       requires_grad=False).to(args.device)
+    # Maintain one weight vector for BER.
+    class_weights = torch.tensor(
+        1.0 / source_label_distribution, dtype=torch.float, requires_grad=False).to(args.device)
+
+    for epoch in range(1, args.epochs + 1):
+        start_time_test = time.time()
+        if epoch % decay_epoch == 0:
+            for param_group in optimizer.param_groups:
+                param_group["lr"] = param_group["lr"] * decay_frac
+        test(args, epoch, model, test_samples, test_labels, start_time_test, out_log_file, name='Target test')
+        train(args, model, ad_net, source_samples,
+              source_labels, target_samples, target_labels,
+              optimizer, optimizer_ad, epoch, start_epoch, args.method, source_label_distribution, out_wei_file, cov_mat, pseudo_target_label, class_weights, true_weights)
+    test(args, epoch+1, model, test_samples, test_labels, start_time_test, out_log_file, name='Target test')
+    test(args, epoch+1, model, source_samples, source_labels,
+         start_time_test, out_log_file_train, name='Source train')
+
+
+if __name__ == '__main__':
+    main()
--- a/train_image.py
+++ b/train_image.py
@ -0,0 +1,502 @@
+import argparse
+import numpy as np
+import os
+import os.path as osp
+import pickle
+import scipy.stats
+import sys
+import time
+import torch
+import torch.nn as nn
+import torch.optim as optim
+from torch.autograd import Variable
+from torch.utils.data import DataLoader
+import torch.nn.functional as F
+
+import data_list
+from data_list import ImageList, LoadedImageList, sample_ratios, write_list
+import loss
+import lr_schedule
+import math
+import network
+import pre_process as prep
+import random
+
+
+
+
+def image_classification_test_loaded(test_samples, test_labels, model, test_10crop=True, device='cpu'):
+    with torch.no_grad():
+        test_loss = 0
+        correct = 0
+        if test_10crop:
+            len_test = test_labels[0].shape[0]
+            for i in range(len_test):
+                outputs = []
+                for j in range(10):
+                    data, target = test_samples[j][i, :, :, :].unsqueeze(0), test_labels[j][i].unsqueeze(0)
+                    _, output = model(data)
+                    test_loss += nn.CrossEntropyLoss()(output, target).item()
+                    outputs.append(nn.Softmax(dim=1)(output))
+                outputs = sum(outputs)
+                pred = torch.max(outputs, 1)[1]
+                correct += pred.eq(target.data.cpu().view_as(pred)).sum().item()
+        else:
+            len_test = test_labels.shape[0]
+            bs = 72
+            for i in range(int(len_test / bs)):
+                data, target = torch.Tensor(test_samples[bs*i:bs*(i+1), :, :, :]).to(config["device"]), test_labels[bs*i:bs*(i+1)]
+                _, output = model(data)
+                test_loss += nn.CrossEntropyLoss()(output, target).item()
+                pred = torch.max(output, 1)[1]
+                correct += pred.eq(target.data.view_as(pred)).sum().item()
+            # Last test samples
+            data, target = torch.Tensor(test_samples[bs*(i+1):, :, :, :]).to(config["device"]), test_labels[bs*(i+1):]
+            _, output = model(data)
+            test_loss += nn.CrossEntropyLoss()(output, target).item()
+            pred = torch.max(output, 1)[1]
+            correct += pred.eq(target.data.view_as(pred)).sum().item()
+    accuracy = correct / len_test
+    test_loss /= len_test * 10
+    return accuracy
+
+
+def train(config):
+
+    ## Define start time
+    start_time = time.time()
+
+    ## set pre-process
+    prep_dict = {}
+    prep_config = config["prep"]
+    prep_dict["source"] = prep.image_train(**config["prep"]['params'])
+    prep_dict["target"] = prep.image_train(**config["prep"]['params'])
+    if prep_config["test_10crop"]:
+        prep_dict["test"] = prep.image_test_10crop(**config["prep"]['params'])
+    else:
+        prep_dict["test"] = prep.image_test(**config["prep"]['params'])
+
+    ## prepare data
+    print("Preparing data", flush=True)
+    dsets = {}
+    dset_loaders = {}
+    data_config = config["data"]
+    train_bs = data_config["source"]["batch_size"]
+    test_bs = data_config["test"]["batch_size"]
+    root_folder = data_config["root_folder"]
+    dsets["source"] = ImageList(open(osp.join(root_folder, data_config["source"]["list_path"])).readlines(), \
+                                transform=prep_dict["source"], root_folder=root_folder, ratios=config["ratios_source"])
+    dset_loaders["source"] = DataLoader(dsets["source"], batch_size=train_bs, \
+            shuffle=True, num_workers=4, drop_last=True)
+    dsets["target"] = ImageList(open(osp.join(root_folder, data_config["target"]["list_path"])).readlines(), \
+                                transform=prep_dict["target"], root_folder=root_folder, ratios=config["ratios_target"])
+    dset_loaders["target"] = DataLoader(dsets["target"], batch_size=train_bs, \
+            shuffle=True, num_workers=4, drop_last=True)
+
+    if prep_config["test_10crop"]:
+        dsets["test"] = [ImageList(open(osp.join(root_folder, data_config["test"]["list_path"])).readlines(),
+                                    transform=prep_dict["test"][i], root_folder=root_folder, ratios=config["ratios_test"]) for i in range(10)]
+        dset_loaders["test"] = [DataLoader(dset, batch_size=test_bs, \
+                            shuffle=False, num_workers=4) for dset in dsets['test']]
+    else:
+        dsets["test"] = ImageList(open(osp.join(root_folder, data_config["test"]["list_path"])).readlines(),
+                                  transform=prep_dict["test"], root_folder=root_folder, ratios=config["ratios_test"])
+        dset_loaders["test"] = DataLoader(dsets["test"], batch_size=test_bs, \
+                                shuffle=False, num_workers=4)
+
+    test_path = os.path.join(root_folder, data_config["test"]["dataset_path"])
+    if os.path.exists(test_path):
+        print('Found existing dataset for test', flush=True)
+        with open(test_path, 'rb') as f:
+            [test_samples, test_labels] = pickle.load(f)
+            test_labels = torch.LongTensor(test_labels).to(config["device"])
+    else:
+        print('Missing test dataset', flush=True)
+        print('Building dataset for test and writing to {}'.format(
+            test_path), flush=True)
+        if prep_config["test_10crop"]:
+            dsets_test = [ImageList(open(osp.join(root_folder, data_config["test"]["list_path"])).readlines(),
+                                    transform=prep_dict["test"][i], root_folder=root_folder) for i in range(10)]
+            loaded_dsets_test = [LoadedImageList(
+                dset_test) for dset_test in dsets_test]
+            test_samples, test_labels = [loaded_dset_test.samples.numpy() for loaded_dset_test in loaded_dsets_test], \
+                                        [loaded_dset_test.targets.numpy()
+                                         for loaded_dset_test in loaded_dsets_test]
+            with open(test_path, 'wb') as f:
+                pickle.dump([test_samples, test_labels], f)
+        else:
+            dset_test = ImageList(open(osp.join(root_folder, data_config["test"]["list_path"])).readlines(),
+                                  transform=prep_dict["test"], root_folder=root_folder, ratios=config['ratios_test'])
+            loaded_dset_test = LoadedImageList(dset_test)
+            test_samples, test_labels = loaded_dset_test.samples.numpy(), loaded_dset_test.targets.numpy()
+            with open(test_path, 'wb') as f:
+                pickle.dump([test_samples, test_labels], f)
+
+    class_num = config["network"]["params"]["class_num"]
+    test_samples, test_labels = sample_ratios(
+        test_samples, test_labels, config['ratios_test'])
+
+    # compute labels distribution on the source and target domain
+    source_label_distribution = np.zeros((class_num))
+    for img in dsets["source"].imgs:
+        source_label_distribution[img[1]] += 1
+    print("Total source samples: {}".format(np.sum(source_label_distribution)), flush=True)
+    print("Source samples per class: {}".format(source_label_distribution), flush=True)
+    source_label_distribution /= np.sum(source_label_distribution)
+    print("Source label distribution: {}".format(source_label_distribution), flush=True)
+    target_label_distribution = np.zeros((class_num))
+    for img in dsets["target"].imgs:
+        target_label_distribution[img[1]] += 1
+    print("Total target samples: {}".format(
+        np.sum(target_label_distribution)), flush=True)
+    print("Target samples per class: {}".format(target_label_distribution), flush=True)
+    target_label_distribution /= np.sum(target_label_distribution)
+    print("Target label distribution: {}".format(target_label_distribution), flush=True)
+    mixture = (source_label_distribution + target_label_distribution) / 2
+    jsd = (scipy.stats.entropy(source_label_distribution, qk=mixture) \
+            + scipy.stats.entropy(target_label_distribution, qk=mixture)) / 2
+    print("JSD : {}".format(jsd), flush=True)
+
+    test_label_distribution = np.zeros((class_num))
+    for img in test_labels:
+        test_label_distribution[int(img.item())] += 1
+    print("Test samples per class: {}".format(test_label_distribution), flush=True)
+    test_label_distribution /= np.sum(test_label_distribution)
+    print("Test label distribution: {}".format(test_label_distribution), flush=True)
+    write_list(config["out_wei_file"], [round(x, 4) for x in test_label_distribution])
+    write_list(config["out_wei_file"], [round(x, 4) for x in source_label_distribution])
+    write_list(config["out_wei_file"], [round(x, 4) for x in target_label_distribution])
+    true_weights = torch.tensor(
+        target_label_distribution / source_label_distribution, dtype=torch.float, requires_grad=False)[:, None].to(config["device"])
+    print("True weights : {}".format(true_weights[:, 0].cpu().numpy()))
+    config["out_wei_file"].write(str(jsd) + "\n")
+
+    ## set base network
+    net_config = config["network"]
+    base_network = net_config["name"](**net_config["params"])
+    base_network = base_network.to(config["device"])
+
+    ## add additional network for some methods
+    if config["loss"]["random"]:
+        random_layer = network.RandomLayer([base_network.output_num(), class_num], config["loss"]["random_dim"])
+        ad_net = network.AdversarialNetwork(config["loss"]["random_dim"], 1024)
+    else:
+        random_layer = None
+        if 'CDAN' in config['method']:
+            ad_net = network.AdversarialNetwork(base_network.output_num() * class_num, 1024)
+        else:
+            ad_net = network.AdversarialNetwork(base_network.output_num(), 1024)
+    if config["loss"]["random"]:
+        random_layer.to(config["device"])
+    ad_net = ad_net.to(config["device"])
+    parameter_list = ad_net.get_parameters() + base_network.get_parameters()
+    parameter_list[-1]["lr_mult"] = config["lr_mult_im"]
+
+    ## set optimizer
+    optimizer_config = config["optimizer"]
+    optimizer = optimizer_config["type"](parameter_list, \
+                    **(optimizer_config["optim_params"]))
+    param_lr = []
+    for param_group in optimizer.param_groups:
+        param_lr.append(param_group["lr"])
+    schedule_param = optimizer_config["lr_param"]
+    lr_scheduler = lr_schedule.schedule_dict[optimizer_config["lr_type"]]
+
+    # Maintain two quantities for the QP.
+    cov_mat = torch.tensor(np.zeros((class_num, class_num), dtype=np.float32),
+                           requires_grad=False).to(config["device"])
+    pseudo_target_label = torch.tensor(np.zeros((class_num, 1), dtype=np.float32),
+                                       requires_grad=False).to(config["device"])
+    # Maintain one weight vector for BER.
+    class_weights = torch.tensor(
+        1.0 / source_label_distribution, dtype=torch.float, requires_grad=False).to(config["device"])
+
+    gpus = config['gpu'].split(',')
+    if len(gpus) > 1:
+        ad_net = nn.DataParallel(ad_net, device_ids=[int(i) for i in gpus])
+        base_network = nn.DataParallel(base_network, device_ids=[int(i) for i in gpus])
+
+    ## train
+    len_train_source = len(dset_loaders["source"])
+    len_train_target = len(dset_loaders["target"])
+    transfer_loss_value = classifier_loss_value = total_loss_value = 0.0
+    best_acc = 0.0
+
+    print("Preparations done in {:.0f} seconds".format(time.time() - start_time), flush=True)
+    print("Starting training for {} iterations using method {}".format(config["num_iterations"], config['method']), flush=True)
+    start_time_test = start_time = time.time()
+    for i in range(config["num_iterations"]):
+        if i % config["test_interval"] == config["test_interval"] - 1:
+            base_network.train(False)
+            temp_acc = image_classification_test_loaded(test_samples, test_labels, base_network, test_10crop=prep_config["test_10crop"])
+            temp_model = nn.Sequential(base_network)
+            if temp_acc > best_acc:
+                best_acc = temp_acc
+            log_str = "  iter: {:05d}, sec: {:.0f}, class: {:.5f}, da: {:.5f}, precision: {:.5f}".format(
+                i, time.time() - start_time_test, classifier_loss_value, transfer_loss_value, temp_acc)
+            config["out_log_file"].write(log_str+"\n")
+            config["out_log_file"].flush()
+            print(log_str, flush=True)
+            if 'IW' in config['method']:
+                current_weights = [round(x, 4) for x in base_network.im_weights.data.cpu().numpy().flatten()]
+                # write_list(config["out_wei_file"], current_weights)
+                print(current_weights, flush=True)
+            start_time_test = time.time()
+        if i % 500 == -1:
+            print("{} iterations in {} seconds".format(i, time.time() - start_time), flush=True)
+
+        loss_params = config["loss"]
+        ## train one iter
+        base_network.train(True)
+        ad_net.train(True)
+        optimizer = lr_scheduler(optimizer, i, **schedule_param)
+        optimizer.zero_grad()
+
+        t = time.time()
+        if i % len_train_source == 0:
+            iter_source = iter(dset_loaders["source"])
+        if i % len_train_target == 0:
+            iter_target = iter(dset_loaders["target"])
+        inputs_source, label_source = iter_source.next()
+        inputs_target, _ = iter_target.next()
+        inputs_source, inputs_target, label_source = inputs_source.to(config["device"]), inputs_target.to(config["device"]), label_source.to(config["device"])
+        features_source, outputs_source = base_network(inputs_source)
+        features_target, outputs_target = base_network(inputs_target)
+        features = torch.cat((features_source, features_target), dim=0)
+        outputs = torch.cat((outputs_source, outputs_target), dim=0)
+        softmax_out = nn.Softmax(dim=1)(outputs)
+
+        if 'IW' in config['method']:
+            ys_onehot = torch.zeros(train_bs, class_num).to(config["device"])
+            ys_onehot.scatter_(1, label_source.view(-1, 1), 1)
+
+            # Compute weights on source data.
+            if 'ORACLE' in config['method']:
+                weights = torch.mm(ys_onehot, true_weights)
+            else:
+                weights = torch.mm(ys_onehot, base_network.im_weights)
+
+            source_preds, target_preds = outputs[:train_bs], outputs[train_bs:]
+            # Compute the aggregated distribution of pseudo-label on the target domain.
+            pseudo_target_label += torch.sum(
+                F.softmax(target_preds, dim=1), dim=0).view(-1, 1).detach()
+            # Update the covariance matrix on the source domain as well.
+            cov_mat += torch.mm(F.softmax(source_preds,
+                                          dim=1).transpose(1, 0), ys_onehot).detach()
+
+        if config['method'] == 'CDAN-E':
+            classifier_loss = nn.CrossEntropyLoss()(outputs_source, label_source)
+            entropy = loss.Entropy(softmax_out)
+            transfer_loss = loss.CDAN([features, softmax_out], ad_net, entropy, network.calc_coeff(i), random_layer)
+            total_loss = loss_params["trade_off"] * \
+                transfer_loss + classifier_loss
+
+        elif 'IWCDAN-E' in config['method']:
+
+            classifier_loss = torch.mean(
+                nn.CrossEntropyLoss(weight=class_weights, reduction='none')
+                (outputs_source, label_source) * weights) / class_num
+
+            entropy = loss.Entropy(softmax_out)
+            transfer_loss = loss.CDAN(
+                [features, softmax_out], ad_net, entropy, network.calc_coeff(i), random_layer, weights=weights, device=config["device"])
+            total_loss = loss_params["trade_off"] * \
+                transfer_loss + classifier_loss
+
+        elif config['method'] == 'CDAN':
+
+            classifier_loss = nn.CrossEntropyLoss()(outputs_source, label_source)
+            transfer_loss = loss.CDAN([features, softmax_out], ad_net, None, None, random_layer)
+            total_loss = loss_params["trade_off"] * transfer_loss + classifier_loss
+
+        elif 'IWCDAN' in config['method']:
+
+            classifier_loss = torch.mean(
+                nn.CrossEntropyLoss(weight=class_weights, reduction='none')
+                (outputs_source, label_source) * weights) / class_num
+
+            transfer_loss = loss.CDAN([features, softmax_out], ad_net, None, None, random_layer, weights=weights)
+            total_loss = loss_params["trade_off"] * \
+                transfer_loss + classifier_loss
+
+        elif config['method']  == 'DANN':
+            classifier_loss = nn.CrossEntropyLoss()(outputs_source, label_source)
+            transfer_loss = loss.DANN(features, ad_net, config["device"])
+            total_loss = loss_params["trade_off"] * \
+                transfer_loss + classifier_loss
+
+        elif 'IWDAN' in config['method']:
+
+            classifier_loss = torch.mean(
+                nn.CrossEntropyLoss(weight=class_weights, reduction='none')
+                (outputs_source, label_source) * weights) / class_num
+
+            transfer_loss = loss.IWDAN(features, ad_net, weights)
+            total_loss = loss_params["trade_off"] * \
+                transfer_loss + classifier_loss
+
+        elif config['method'] == 'NANN':
+            classifier_loss = nn.CrossEntropyLoss()(outputs_source, label_source)
+            total_loss = classifier_loss
+        else:
+            raise ValueError('Method cannot be recognized.')
+
+        total_loss.backward()
+        optimizer.step()
+
+        transfer_loss_value = 0 if config['method'] == 'NANN' else transfer_loss.item()
+        classifier_loss_value = classifier_loss.item()
+        total_loss_value = transfer_loss_value + classifier_loss_value
+
+        if ('IW' in config['method']) and i % (config["dataset_mult_iw"] * len_train_source) == config["dataset_mult_iw"] * len_train_source - 1:
+
+            pseudo_target_label /= train_bs * \
+                len_train_source * config["dataset_mult_iw"]
+            cov_mat /= train_bs * len_train_source * config["dataset_mult_iw"]
+            print(i, np.sum(cov_mat.cpu().detach().numpy()), train_bs * len_train_source)
+
+            # Recompute the importance weight by solving a QP.
+            base_network.im_weights_update(source_label_distribution,
+                                           pseudo_target_label.cpu().detach().numpy(),
+                                           cov_mat.cpu().detach().numpy(),
+                                           config["device"])
+            current_weights = [
+                round(x, 4) for x in base_network.im_weights.data.cpu().numpy().flatten()]
+            write_list(config["out_wei_file"], [np.linalg.norm(
+                current_weights - true_weights.cpu().numpy().flatten())] + current_weights)
+            print(np.linalg.norm(current_weights -
+                                true_weights.cpu().numpy().flatten()), current_weights)
+
+            cov_mat[:] = 0.0
+            pseudo_target_label[:] = 0.0
+
+    return best_acc
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Conditional Domain Adversarial Network')
+    parser.add_argument('method', type=str, choices=[
+                        'NANN', 'DANN', 'IWDAN', 'IWDANORACLE', 'CDAN', 'IWCDAN', 'IWCDANORACLE', 'CDAN-E', 'IWCDAN-E', 'IWCDAN-EORACLE'])
+    parser.add_argument('--gpu_id', type=str, nargs='?', default='0', help="device id to run")
+    parser.add_argument('--net', type=str, default='ResNet50', choices=["ResNet18", "ResNet34", "ResNet50", "ResNet101", "ResNet152", "VGG11", "VGG13", "VGG16", "VGG19", "VGG11BN", "VGG13BN", "VGG16BN", "VGG19BN", "AlexNet"], help="Network type. Only tested with ResNet50")
+    parser.add_argument('--dset', type=str, default='office-31', choices=['office-31', 'visda', 'office-home'], help="The dataset or source dataset used")
+    parser.add_argument('--s_dset_file', type=str, default='amazon_list.txt', help="The source dataset path list")
+    parser.add_argument('--t_dset_file', type=str, default='webcam_list.txt', help="The target dataset path list")
+    parser.add_argument('--test_interval', type=int, default=500, help="interval of two continuous test phase")
+    parser.add_argument('--snapshot_interval', type=int, default=10000, help="interval of two continuous output model")
+    parser.add_argument('--output_dir', type=str, default='results', help="output directory")
+    parser.add_argument('--root_folder', type=str, default=None, help="The folder containing the datasets")
+    parser.add_argument('--lr', type=float, default=0.001,
+                        help="learning rate")
+    parser.add_argument('--trade_off', type=float, default=1.0, help="factor for dann")
+    parser.add_argument('--random', type=bool, default=False, help="whether use random projection")
+    parser.add_argument('--seed', type=int, default='42', help="Random seed")
+    parser.add_argument('--lr_mult_im', type=int, default='1', help="Multiplicative factor for IM")
+    parser.add_argument('--dataset_mult_iw', type=int, default='0', help="Frequency of weight updates in multiples of the dataset. Default: 1 for digits and visda, 15 for office datasets")
+    parser.add_argument('--num_iterations', type=int, default='100000', help="Number of batch updates")
+    parser.add_argument('--ratio', type=int, default=0, help='ratio option. If 0 original dataset, if 1, only 30% of samples in the first half of the classes are considered')
+    parser.add_argument('--ma', type=float, default=0.5,
+                        help='weight for the moving average of iw')
+    args = parser.parse_args()
+
+    if args.root_folder is None:
+        args.root_folder = 'data/{}/'.format(args.dset)
+
+    if args.s_dset_file != args.t_dset_file:
+        # Set GPU ID
+        os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu_id
+
+        # Set random number seed.
+        np.random.seed(args.seed)
+        torch.manual_seed(args.seed)
+
+        # train config
+        config = {}
+        config['method'] = args.method
+        config["gpu"] = args.gpu_id
+        config["device"] = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        config["num_iterations"] = args.num_iterations
+        config["test_interval"] = args.test_interval
+        config["snapshot_interval"] = args.snapshot_interval
+        config["output_for_test"] = True
+        config["output_path"] = args.output_dir
+        if not osp.exists(config["output_path"]):
+            os.system('mkdir -p '+ config["output_path"])
+        config["out_log_file"] = open(osp.join(config["output_path"], "log.txt"), "w")
+        config["out_wei_file"] = open(osp.join(config["output_path"], "log_weights.txt"), "w")
+        if not osp.exists(config["output_path"]):
+            os.mkdir(config["output_path"])
+
+        config["prep"] = {"test_10crop":False, 'params':{"resize_size":256, "crop_size":224, 'alexnet':False}}
+        config["loss"] = {"trade_off":args.trade_off}
+        if "AlexNet" in args.net:
+            config["prep"]['params']['alexnet'] = True
+            config["prep"]['params']['crop_size'] = 227
+            config["network"] = {"name":network.AlexNetFc, \
+                "params":{"use_bottleneck":True, "bottleneck_dim":256, "new_cls":True, "ma": args.ma} }
+        elif "ResNet" in args.net:
+            config["network"] = {"name":network.ResNetFc, \
+                "params":{"resnet_name":args.net, "use_bottleneck":True, "bottleneck_dim":256, "new_cls":True, "ma": args.ma} }
+        elif "VGG" in args.net:
+            config["network"] = {"name":network.VGGFc, \
+                "params":{"vgg_name":args.net, "use_bottleneck":True, "bottleneck_dim":256, "new_cls":True, "ma": args.ma} }
+        config["loss"]["random"] = args.random
+        config["loss"]["random_dim"] = 1024
+
+        config["optimizer"] = {"type":optim.SGD, "optim_params":{'lr':args.lr, "momentum":0.9, \
+                            "weight_decay":0.0005, "nesterov":True}, "lr_type":"inv", \
+                            "lr_param":{"lr":args.lr, "gamma":0.001, "power":0.75} }
+
+        config["dataset"] = args.dset
+        config["data"] = {"source":{"list_path":args.s_dset_file, "batch_size":36}, \
+                          "target":{"list_path":args.t_dset_file, "batch_size":36}, \
+                          "test": {"list_path": args.t_dset_file, "dataset_path": "{}_test.pkl".format(args.t_dset_file), "batch_size": 4},
+                          "root_folder":args.root_folder}
+
+        config["lr_mult_im"] = args.lr_mult_im
+        if config["dataset"] == "office-31":
+            if ("amazon" in args.s_dset_file and "webcam" in args.t_dset_file) or \
+            ("webcam" in args.s_dset_file and "dslr" in args.t_dset_file) or \
+            ("webcam" in args.s_dset_file and "amazon" in args.t_dset_file) or \
+            ("dslr" in args.s_dset_file and "amazon" in args.t_dset_file):
+                config["optimizer"]["lr_param"]["lr"] = 0.001 # optimal parameters
+            elif ("amazon" in args.s_dset_file and "dslr" in args.t_dset_file) or \
+                ("dslr" in args.s_dset_file and "webcam" in args.t_dset_file):
+                config["optimizer"]["lr_param"]["lr"] = 0.0003 # optimal parameters
+            config["network"]["params"]["class_num"] = 31
+            config["ratios_source"] = [1] * 31
+            if args.ratio == 1:
+                config["ratios_source"] = [0.3] * 15 + [1] * 16
+            config["ratios_target"] = [1] * 31
+            if args.dataset_mult_iw == 0:
+                args.dataset_mult_iw = 15
+        elif config["dataset"] == "visda":
+            config["optimizer"]["lr_param"]["lr"] = 0.001 # optimal parameters
+            config["network"]["params"]["class_num"] = 12
+            config["ratios_source"] = [1] * 12
+            if args.ratio == 1:
+                config["ratios_source"] = [0.3] * 6 + [1] * 6
+            config["ratios_target"] = [1] * 12
+            if args.dataset_mult_iw == 0:
+                args.dataset_mult_iw = 1
+        elif config["dataset"] == "office-home":
+            config["optimizer"]["lr_param"]["lr"] = 0.001 # optimal parameters
+            config["network"]["params"]["class_num"] = 65
+            config["ratios_source"] = [1] * 65
+            if args.ratio == 1:
+                config["ratios_source"] = [0.3] * 32 + [1] * 33
+            config["ratios_target"] = [1] * 65
+            if args.dataset_mult_iw == 0:
+                args.dataset_mult_iw = 15
+        else:
+            raise ValueError('Dataset cannot be recognized. Please define your own dataset here.')
+
+        config["dataset_mult_iw"] = args.dataset_mult_iw
+        config["ratios_test"] = config["ratios_target"]
+        config["out_log_file"].write(str(config) + "\n")
+        config["out_log_file"].flush()
+
+        print("-" * 50, flush=True)
+        print("\nRunning {} on the {} dataset with source {} and target {} and trade off {}\n".format(args.method, args.dset,args.s_dset_file, args.t_dset_file, args.trade_off), flush=True )
+        print("-" * 50, flush=True)
+        train(config)
--- a/train_mmd.py
+++ b/train_mmd.py
@ -0,0 +1,415 @@
+import argparse
+import os, sys
+import os.path as osp
+import pickle
+import numpy as np
+import scipy.stats
+import torch
+import torch.nn as nn
+import torch.optim as optim
+import network
+import loss
+import pre_process as prep
+from torch.utils.data import DataLoader
+import torch.nn.functional as F
+import lr_schedule
+from data_list import ImageList, LoadedImageList, sample_ratios, image_classification_test_loaded, write_list
+from torch.autograd import Variable
+
+
+optim_dict = {"SGD": optim.SGD}
+
+
+class AverageMeter(object):
+    """Computes and stores the average and current value"""
+    def __init__(self):
+        self.reset()
+
+    def reset(self):
+        self.val = 0
+        self.avg = 0
+        self.sum = 0
+        self.count = 0
+
+    def update(self, val, n=1):
+        self.val = val
+        self.sum += val * n
+        self.count += n
+        self.avg = self.sum // self.count
+
+
+def transfer_classification(config):
+
+    use_gpu = torch.cuda.is_available()
+    device = 'cuda' if use_gpu else 'cpu'
+
+    ## set pre-process
+    prep_dict = {}
+    prep_config = config["prep"]
+    prep_dict["source"] = prep.image_train(**prep_config['params'])
+    prep_dict["target"] = prep.image_train(**prep_config['params'])
+    if prep_config["test_10crop"]:
+        prep_dict["test"] = prep.image_test_10crop(**prep_config['params'])
+    else:
+        prep_dict["test"] = prep.image_test(**prep_config['params'])
+
+    ## set loss
+    class_criterion = nn.CrossEntropyLoss()
+    loss_config = config["loss"]
+    transfer_criterion = loss.loss_dict[loss_config["name"]]
+    if "params" not in loss_config:
+        loss_config["params"] = {}
+
+    ## prepare data
+    print("Preparing data", flush=True)
+    dsets = {}
+    dset_loaders = {}
+    data_config = config["data"]
+    train_bs = data_config["source"]["batch_size"]
+    test_bs = data_config["test"]["batch_size"]
+    root_folder = data_config["root_folder"]
+    dsets["source"] = ImageList(open(osp.join(root_folder, data_config["source"]["list_path"])).readlines(),
+                                transform=prep_dict["source"], root_folder=root_folder, ratios=config["ratios_source"], mode=prep_config['mode'])
+    dset_loaders["source"] = DataLoader(dsets["source"], batch_size=train_bs,
+                                        shuffle=True, num_workers=4, drop_last=True)
+    dsets["target"] = ImageList(open(osp.join(root_folder, data_config["target"]["list_path"])).readlines(),
+                                transform=prep_dict["target"], root_folder=root_folder, ratios=config["ratios_target"], mode=prep_config['mode'])
+    dset_loaders["target"] = DataLoader(dsets["target"], batch_size=train_bs,
+                                        shuffle=True, num_workers=4, drop_last=True)
+
+    if prep_config["test_10crop"]:
+        dsets["test"] = [ImageList(open(osp.join(root_folder, data_config["test"]["list_path"])).readlines(),
+                                   transform=prep_dict["test"][i], root_folder=root_folder, ratios=config["ratios_test"], mode=prep_config['mode']) for i in range(10)]
+        dset_loaders["test"] = [DataLoader(dset, batch_size=test_bs,
+                                           shuffle=False, num_workers=4) for dset in dsets['test']]
+    else:
+        dsets["test"] = ImageList(open(osp.join(root_folder, data_config["test"]["list_path"])).readlines(),
+                                  transform=prep_dict["test"], root_folder=root_folder, ratios=config["ratios_test"], mode=prep_config['mode'])
+        dset_loaders["test"] = DataLoader(dsets["test"], batch_size=test_bs,
+                                          shuffle=False, num_workers=4)
+
+    test_path = os.path.join(root_folder, data_config["test"]["dataset_path"])
+    if os.path.exists(test_path):
+        print('Found existing dataset for test', flush=True)
+        with open(test_path, 'rb') as f:
+            [test_samples, test_labels] = pickle.load(f)
+            test_labels = torch.LongTensor(test_labels).to(device)
+    else:
+        print('Missing test dataset', flush=True)
+        print('Building dataset for test and writing to {}'.format(test_path), flush=True)
+        if prep_config["test_10crop"]:
+            dsets_test = [ImageList(open(osp.join(root_folder, data_config["test"]["list_path"])).readlines(),
+                                    transform=prep_dict["test"][i], root_folder=root_folder) for i in range(10)]
+            loaded_dsets_test = [LoadedImageList(
+                dset_test) for dset_test in dsets_test]
+            test_samples, test_labels = [loaded_dset_test.samples.numpy() for loaded_dset_test in loaded_dsets_test], \
+                                        [loaded_dset_test.targets.numpy()
+                                         for loaded_dset_test in loaded_dsets_test]
+            with open(test_path, 'wb') as f:
+                pickle.dump([test_samples, test_labels], f)
+        else:
+            dset_test = ImageList(open(osp.join(root_folder, data_config["test"]["list_path"])).readlines(),
+                                  transform=prep_dict["test"], root_folder=root_folder, ratios=config['ratios_test'])
+            loaded_dset_test = LoadedImageList(dset_test)
+            test_samples, test_labels = loaded_dset_test.samples.numpy(
+            ), loaded_dset_test.targets.numpy()
+            with open(test_path, 'wb') as f:
+                pickle.dump([test_samples, test_labels], f)
+
+    class_num = config["network"]["class_num"]
+    test_samples, test_labels = sample_ratios(
+        test_samples, test_labels, config['ratios_test'])
+
+    ## set base network
+    net_config = config["network"]
+    base_network = network.network_dict[net_config["name"]](**net_config)
+    base_network = base_network.to(device)
+    if net_config["use_bottleneck"]:
+        bottleneck_layer = nn.Linear(base_network.output_num(), net_config["bottleneck_dim"]).to(device)
+        classifier_layer = nn.Linear(bottleneck_layer.out_features, class_num)
+    else:
+        classifier_layer = nn.Linear(base_network.output_num(), class_num)
+    for param in base_network.parameters():
+        param.requires_grad = False
+
+    classifier_layer = classifier_layer.to(device)
+
+    ## initialization
+    if net_config["use_bottleneck"]:
+        bottleneck_layer.weight.data.normal_(0, 0.005)
+        bottleneck_layer.bias.data.fill_(0.1)
+        bottleneck_layer = nn.Sequential(bottleneck_layer, nn.ReLU(), nn.Dropout(0.5))
+    classifier_layer.weight.data.normal_(0, 0.01)
+    classifier_layer.bias.data.fill_(0.0)
+
+    ## collect parameters
+    if net_config["use_bottleneck"]:
+        parameter_list = [{"params":bottleneck_layer.parameters(), "lr":10}, {"params":classifier_layer.parameters(), "lr":10}]
+
+    else:
+        parameter_list = [{"params":classifier_layer.parameters(), "lr":10}]
+
+    # compute labels distribution on the source and target domain
+    source_label_distribution = np.zeros((class_num))
+    for img in dsets["source"].imgs:
+        source_label_distribution[img[1]] += 1
+    print("Total source samples: {}".format(
+        np.sum(source_label_distribution)), flush=True)
+    print("Source samples per class: {}".format(
+        source_label_distribution), flush=True)
+    source_label_distribution /= np.sum(source_label_distribution)
+    print("Source label distribution: {}".format(
+        source_label_distribution), flush=True)
+    target_label_distribution = np.zeros((class_num))
+    for img in dsets["target"].imgs:
+        target_label_distribution[img[1]] += 1
+    print("Total target samples: {}".format(
+        np.sum(target_label_distribution)), flush=True)
+    print("Target samples per class: {}".format(
+        target_label_distribution), flush=True)
+    target_label_distribution /= np.sum(target_label_distribution)
+    print("Target label distribution: {}".format(
+        target_label_distribution), flush=True)
+    mixture = (source_label_distribution + target_label_distribution) / 2
+    jsd = (scipy.stats.entropy(source_label_distribution, qk=mixture)
+           + scipy.stats.entropy(target_label_distribution, qk=mixture)) / 2
+    print("JSD : {}".format(jsd), flush=True)
+    true_weights = torch.tensor(
+        target_label_distribution / source_label_distribution, dtype=torch.float, requires_grad=False)[:, None].to(device)
+    write_list(config["out_wei_file"], [round(x, 4) for x in source_label_distribution])
+    write_list(config["out_wei_file"], [round(x, 4) for x in target_label_distribution])
+    print("True weights : {}".format(true_weights[:, 0].cpu().numpy()))
+    config["out_wei_file"].write(str(jsd) + "\n")
+
+    ## set optimizer
+    optimizer_config = config["optimizer"]
+    optimizer = optim_dict[optimizer_config["type"]](parameter_list, **(optimizer_config["optim_params"]))
+    param_lr = []
+    for param_group in optimizer.param_groups:
+        param_lr.append(param_group["lr"])
+    schedule_param = optimizer_config["lr_param"]
+    lr_scheduler = lr_schedule.schedule_dict[optimizer_config["lr_type"]]
+
+    # Maintain two quantities for the QP.
+    cov_mat = torch.tensor(np.zeros((class_num, class_num), dtype=np.float32),
+                           requires_grad=False).to(device)
+    pseudo_target_label = torch.tensor(np.zeros((class_num, 1), dtype=np.float32),
+                                       requires_grad=False).to(device)
+    # Maintain one weight vector for BER.
+    class_weights = torch.tensor(
+        1.0 / source_label_distribution, dtype=torch.float, requires_grad=False).to(device)
+
+    ## train
+    len_train_source = len(dset_loaders["source"])
+    len_train_target = len(dset_loaders["target"])
+    mmd_meter = AverageMeter()
+    for i in range(config["num_iterations"]):
+        ## test in the train
+        if i % config["test_interval"] == 1:
+            base_network.train(False)
+            classifier_layer.train(False)
+            if net_config["use_bottleneck"]:
+                bottleneck_layer.train(False)
+                test_acc = image_classification_test_loaded(
+                    test_samples, test_labels, nn.Sequential(
+                        base_network, bottleneck_layer, classifier_layer), test_10crop=prep_config["test_10crop"], device=device)
+            else:
+                test_acc = image_classification_test_loaded(
+                    test_samples, test_labels, nn.Sequential(
+                        base_network, classifier_layer), test_10crop=prep_config["test_10crop"], device=device)
+
+            log_str = 'Iter: %d, mmd = %.4f, test_acc = %.3f' % (
+                i, mmd_meter.avg, test_acc)
+            print(log_str)
+            config["out_log_file"].write(log_str+"\n")
+            config["out_log_file"].flush()
+            mmd_meter.reset()
+
+        ## train one iter
+        if net_config["use_bottleneck"]:
+            bottleneck_layer.train(True)
+        classifier_layer.train(True)
+        optimizer = lr_scheduler(param_lr, optimizer, i, **schedule_param)
+        optimizer.zero_grad()
+        if i % len_train_source == 0:
+            iter_source = iter(dset_loaders["source"])
+        if i % len_train_target == 0:
+            iter_target = iter(dset_loaders["target"])
+        inputs_source, labels_source = iter_source.next()
+        inputs_target, labels_target = iter_target.next()
+
+        inputs_source, inputs_target, labels_source = Variable(inputs_source).to(device), Variable(inputs_target).to(device), Variable(labels_source).to(device)
+
+        inputs = torch.cat((inputs_source, inputs_target), dim=0)
+
+        features = base_network(inputs)
+        if net_config["use_bottleneck"]:
+            features = bottleneck_layer(features)
+
+        outputs = classifier_layer(features)
+
+        if 'IW' in loss_config["name"]:
+            ys_onehot = torch.zeros(train_bs, class_num).to(device)
+            ys_onehot.scatter_(1, labels_source.view(-1, 1), 1)
+
+            # Compute weights on source data.
+            if 'ORACLE' in loss_config["name"]:
+                weights = torch.mm(ys_onehot, true_weights)
+            else:
+                weights = torch.mm(ys_onehot, base_network.im_weights)
+
+            source_preds, target_preds = outputs[:train_bs], outputs[train_bs:]
+            # Compute the aggregated distribution of pseudo-label on the target domain.
+            pseudo_target_label += torch.sum(F.softmax(target_preds, dim=1), dim=0).view(-1, 1).detach()
+            # Update the covariance matrix on the source domain as well.
+            cov_mat += torch.mm(F.softmax(source_preds,
+                                          dim=1).transpose(1, 0), ys_onehot).detach()
+
+            classifier_loss = torch.mean(
+                nn.CrossEntropyLoss(weight=class_weights, reduction='none')
+                (outputs.narrow(0, 0, inputs.size(0)//2), labels_source) * weights) / class_num
+        else:
+            classifier_loss = class_criterion(
+                outputs.narrow(0, 0, inputs.size(0)//2), labels_source)
+
+        ## switch between different transfer loss
+        if loss_config["name"] == "DAN" or loss_config["name"] == "DAN_Linear":
+            transfer_loss = transfer_criterion(features.narrow(0, 0, features.size(0)//2), features.narrow(0, features.size(0)//2, features.size(0)//2), **loss_config["params"])
+        elif loss_config["name"] == "JAN" or loss_config["name"] == "JAN_Linear":
+            softmax_out = nn.Softmax(dim=1)(outputs)
+            transfer_loss = transfer_criterion([features.narrow(0, 0, features.size(0)//2), softmax_out.narrow(0, 0, softmax_out.size(0)//2)], [features.narrow(0, features.size(0)//2, features.size(0)//2), softmax_out.narrow(0, softmax_out.size(0)//2, softmax_out.size(0)//2)], **loss_config["params"])
+        elif "IWJAN" in loss_config["name"]:
+            softmax_out = nn.Softmax(dim=1)(outputs)
+            transfer_loss = transfer_criterion([features.narrow(0, 0, features.size(0)//2), softmax_out.narrow(0, 0, softmax_out.size(0)//2)], [features.narrow(0, features.size(0)//2, features.size(0)//2), softmax_out.narrow(0, softmax_out.size(0)//2, softmax_out.size(0)//2)], weights=weights, **loss_config["params"])
+
+        mmd_meter.update(transfer_loss.item(), inputs_source.size(0))
+        total_loss = loss_config["trade_off"] * transfer_loss + classifier_loss
+        total_loss.backward()
+        optimizer.step()
+
+        if ('IW' in loss_config["name"]) and i % (config["dataset_mult_iw"] * len_train_source) == config["dataset_mult_iw"] * len_train_source - 1:
+
+            if i > config["dataset_mult_iw"] * len_train_source - 1:
+                pseudo_target_label /= train_bs * \
+                    len_train_source * config["dataset_mult_iw"]
+                cov_mat /= train_bs * len_train_source * config["dataset_mult_iw"]
+                print(i, np.sum(cov_mat.cpu().detach().numpy()),
+                    train_bs * len_train_source)
+
+                # Recompute the importance weight by solving a QP.
+                base_network.im_weights_update(source_label_distribution,
+                                               pseudo_target_label.cpu().detach().numpy(),
+                                               cov_mat.cpu().detach().numpy(),
+                                               device)
+
+            current_weights = [
+                round(x, 4) for x in base_network.im_weights.data.cpu().numpy().flatten()]
+            write_list(config["out_wei_file"], [np.linalg.norm(
+                current_weights - true_weights.cpu().numpy().flatten())] + current_weights)
+            print(np.linalg.norm(current_weights -
+                                true_weights.cpu().numpy().flatten()), current_weights)
+
+            cov_mat[:] = 0.0
+            pseudo_target_label[:] = 0.0
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description='Transfer Learning')
+    parser.add_argument('method', type=str, help="loss name",
+                        choices=['JAN', 'IWJAN', 'IWJANORACLE', 'JAN_Linear', 'DAN', 'DAN_Linear', 'IWJAN_Linear', 'IWJANORACLE_Linear'])
+    parser.add_argument('--gpu_id', type=str, nargs='?', default='0', help="device id to run")
+    parser.add_argument('--dset', type=str, default='office', choices=[
+                        'office-31', 'visda', 'office-home'], help="The dataset or source dataset used")
+    parser.add_argument('--s_dset_file', type=str, nargs='?',
+                        default='train_list.txt', help="source data")
+    parser.add_argument('--t_dset_file', type=str, nargs='?',
+                        default='validation_list.txt', help="target data")
+    parser.add_argument('--trade_off', type=float, nargs='?', default=1.0, help="trade_off")
+    parser.add_argument('--output_dir', type=str, default='results', help="output directory")
+    parser.add_argument('--root_folder', type=str, default=None, help="The folder containing the dataset information")
+    parser.add_argument('--seed', type=int, default='42', help="Random seed")
+    parser.add_argument('--dataset_mult_iw', type=int, default='0',
+                        help="Frequency of weight updates in multiples of the dataset")
+    parser.add_argument('--ratio', type=int, default=0, help='ratio option')
+    parser.add_argument('--ma', type=float, default=0.5, help='weight for the moving average of iw')
+    args = parser.parse_args()
+    os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu_id
+
+    if args.root_folder is None:
+        args.root_folder = 'data/{}/'.format(args.dset)
+
+    if args.s_dset_file == args.t_dset_file:
+      sys.exit()
+
+    # Set random number seed.
+    np.random.seed(args.seed)
+    torch.manual_seed(args.seed)
+
+    config = {}
+    config["num_iterations"] = 20000
+    config["test_interval"] = 500
+    config["output_path"] = args.output_dir
+    if not osp.exists(config["output_path"]):
+        os.system('mkdir -p ' + config["output_path"])
+    config["out_log_file"] = open(
+        osp.join(config["output_path"], "log.txt"), "w")
+    config["out_wei_file"] = open(
+        osp.join(config["output_path"], "log_weights.txt"), "w")
+    if not osp.exists(config["output_path"]):
+        os.mkdir(config["output_path"])
+    config["prep"] = {"test_10crop": False, 'params': {"resize_size":256, "crop_size":224, 'alexnet':False}, 'mode':'RGB'}
+    config["loss"] = {"name":args.method, "trade_off":args.trade_off }
+    config["data"] = {"source":{"list_path":args.s_dset_file, "batch_size":36}, \
+                      "target":{"list_path":args.t_dset_file, "batch_size":36}, \
+                      "test": {"list_path": args.t_dset_file, "dataset_path": "{}_test.pkl".format(args.t_dset_file), "batch_size": 4},
+                      "root_folder":args.root_folder}
+    config["network"] = {"name":"ResNet50", "use_bottleneck":True, "bottleneck_dim":256, "ma":args.ma}
+    config["optimizer"] = {"type": "SGD", "optim_params": {"lr": 1.0, "momentum": 0.9, "weight_decay": 0.0005,
+                                                           "nesterov": True}, "lr_type": "inv_mmd", "lr_param": {"gamma": 0.0003, "power": 0.75}}
+
+    config["dataset"] = args.dset
+
+    if config["dataset"] == "office-31":
+        config["optimizer"]["lr_param"]["init_lr"] = 0.0003
+        config["network"]["class_num"] = 31
+        config["ratios_source"] = [1] * 31
+        if args.ratio == 1:
+            config["ratios_source"] = [0.3] * 15 + [1] * 16
+        config["ratios_target"] = [1] * 31
+        config["ratios_test"] = [1] * 31
+        if args.dataset_mult_iw == 0:
+            args.dataset_mult_iw = 15
+    elif config["dataset"] == "visda":
+        config["optimizer"]["lr_param"]["init_lr"] = 0.001
+        config["network"]["class_num"] = 12
+        config["ratios_source"] = [1] * 12
+        if args.ratio == 1:
+            config["ratios_source"] = [0.3] * 6 + [1] * 6
+        config["ratios_target"] = [1] * 12
+        config["ratios_test"] = [1] * 12
+        if args.dataset_mult_iw == 0:
+            args.dataset_mult_iw = 1
+    elif config["dataset"] == "office-home":
+        config["optimizer"]["lr_param"]["init_lr"] = 0.001
+        config["network"]["class_num"] = 65
+        config["ratios_source"] = [1] * 65
+        if args.ratio == 1:
+            config["ratios_source"] = [0.3] * 32 + [1] * 33
+        config["ratios_target"] = [1] * 65
+        config["ratios_test"] = [1] * 65
+        if args.dataset_mult_iw == 0:
+            args.dataset_mult_iw = 15
+    else:
+        raise ValueError(
+            'Dataset cannot be recognized. Please define your own dataset here.')
+
+    config["dataset_mult_iw"] = args.dataset_mult_iw
+    config["out_log_file"].write(str(config) + "\n")
+    config["out_log_file"].flush()
+
+    print("-" * 50, flush=True)
+    print("\nRunning {} on the {} dataset with source {} and target {}\n".format(
+        args.method, args.dset, args.s_dset_file, args.t_dset_file), flush=True)
+    print("-" * 50, flush=True)
+
+    transfer_classification(config)