Update readme

README.md

@@ -32,7 +32,7 @@ If you use our code or models, please [cite](docs/citation.md) our paper.
 
 
 ### What's New?
-
+- 2021.02.03: Release the code for ModelNet40 classification with HRNet.
 - 2020.10.12: Release the initial version of our O-CNN under PyTorch. The code
   has been tested with the [classification task](docs/classification.md#o-cnn-on-pytorch).
 - 2020.08.16: We released our code for [3D unsupervised learning](docs/unsupervised.md).

docs/classification.md

@@ -139,8 +139,9 @@ the data automatically.
 
 ### Train a deep O-CNN-based HRNet
 1. Change the working directory to `tensorflow/data`. Run the following command
-   to store the `points` into one `TFRecords` database. Here we also rotate the
-   upright axis of shapes from `z` axis to `y` axis.
+   to store the `points` into `TFRecords` databases with different ratios of
+   training data. Here we also rotate the upright axis of shapes from `z` axis
+   to `y` axis.
     ```shell
     python cls_modelnet.py --run m40_generate_points_tfrecords_ratios
     ```

docs/unsupervised.md

@@ -4,21 +4,21 @@
 
 1. Download the data and pretrained weights with the following command. 
    ```shell
-   cd  tensorflow
-   python data/midnet_data.py
+   cd  tensorflow/data
+   python midnet_data.py
    ```
 
 2. Run the following command to train the network. 
    ```shell
-   cd  script
+   cd  tensorflow/script
    python run_mid.py --config configs/mid_hrnet_d6.yaml
    ```
    
 ## Finetune on ModelNet40
 
-Follow the instructions [here](classification.md#train-a-deep-o-cnn-based-hrnet)
+Follow the instructions [here](classification.md#prepare-the-point-cloud-for-modelnet40) 
 to download the ModelNet40 and preprocess the data. Make sure you can train the
-HRNet with random initialization.  
+HRNet with random initialization [here](classification.md#train-a-deep-o-cnn-based-hrnet).  
 
 Then run the following script to finetune the network with the pretrained
 weights we provided.  If you would like to finetune the network with your own

tensorflow/script/run_linear_cls_cmd.py

@@ -89,7 +89,7 @@ for i in range(len(ratios)):
   # train the linear classifier
   step_size1 = int(200000 * ratio * muls[i])
   step_size2 = int(100000 * ratio * muls[i])
-  max_iter = int(400000 * ratio * muls[i])
+  max_iter = int(300000 * ratio * muls[i])
   prefix = 'logs/m40/{}/feature_cls/m40_y'.format(alias)
   train_data = '{}_{:.2f}_train_points.tfrecords'.format(prefix, ratio)
   test_data  = '{}_1.00_test_points.tfrecords'.format(prefix)