зеркало из https://github.com/microsoft/EdgeML.git
ef9f8a77f0
* Copying new seedot files * Added batching for compiling large programs without blowing up memory * Updated README.md to FastGRNN example * Fixed copyright messages, included m3 codegen files * Removed unnecessary commented code * Added high level comments and added dlx executables * Fixed formatting of C/C++ files * Adding architecture diagram * Adding tutorial on extending the compiler * Fix typo * Fix Scale Nomenclature and Remove Redundant Function Arguments in MatVec Code * Consolidate c_reference/ and m3/library codes * Revise C and C++ Code To The Agreed Convention * Remove Redundant M3 Library and Add TODO for Memory Management * Re-Index Sparse Model According to Optimal Strategy * Fix Redundant Array Issue * Fix stray spaces and grammatical issues * Make Multi-Dimensional Arrays Uni-Dimensional for M3 * Hacky Fix for Widx and Uidx * Fix Grammar and Remove Unused Code * Make SeeDot Directly Runnable * Remove Stray Spaces * Fix Grammar * Fix Comments * Incorporate Changes from Aayan * Fix x86 pipeline on FirstFitPriority * Optimize x86 pipeline on FirstFitPriority * Add Wno-narrowing flag to x86 codegen * Prettier Printing. (#1) * Replaced the print stagement using logging (error, warning, info and debug levels used). Added progress bar using tqdm package in performSearch function for the 4 stages of exploration. Changes Antlr Check version to 4.9.1. Added requirements.txt * Added more information to the print statment that prints the stage of exploration. * Added log level as a command line argument. Added descriptive error messages. * Reverted parameters to old state * Slight changes to the strings printed by logging * Resolved formatting comments * Updated README.md with comand to change logging level * Converted the log level input arguments to lowercase strings * Minor Improvements * Changes to improve README.md in SeeDot and respective modifications. (#2) * Replaced the print stagement using logging (error, warning, info and debug levels used). Added progress bar using tqdm package in performSearch function for the 4 stages of exploration. Changes Antlr Check version to 4.9.1. Added requirements.txt * Added more information to the print statment that prints the stage of exploration. * Added log level as a command line argument. Added descriptive error messages. * Reverted parameters to old state * Slight changes to the strings printed by logging * Resolved formatting comments * Updated README.md with comand to change logging level * Converted the log level input arguments to lowercase strings * Updated requirements and Readme * Replaced argument version with argument encoding Removed dependency on Results.csv * Readme formatting change * Changed the help strings in argument parser, renamed rnn as fastgrnn * Some more changes to SeeDot-dev.py for the argument parser. Added line that shows that the available fastgrnn model belongs to usps10 dataset. * Added descriptive comments to python files except ONNX and TF folders. * Removed unnecessary comma in launch.json * Corrected default SeeDot values in README.md * Minor changes to improv README.md's readablility. * Added legacy_scales.csv and restored the legacy_scales processor in SeeDot-dev.py * Replaced rnn keyword with fastgrnn in launch.json * Changed all comments to a single format * Added face detection run instructions. Removed lsf flag from README and uppressed it in help. * Replaced maximisingMetric with metric in the main.py, config.py and test.py. Made the respective changes in README.md. * Removed instructions to run face detection * Architecture.md started * Architecture.md iteration 1 completed * Improved line spacing * Removed incorrect line * Replaced architecture.svg * replaced version with encoding * made FastGRNN on usps10 with reduced disagreemts the default for seedot * Points added to architecture.md * Included suggestions from reviewer. * Added instructions to fetch face detection dataset and licenses to newly added files (#3) * Replaced the print stagement using logging (error, warning, info and debug levels used). Added progress bar using tqdm package in performSearch function for the 4 stages of exploration. Changes Antlr Check version to 4.9.1. Added requirements.txt * Added more information to the print statment that prints the stage of exploration. * Added log level as a command line argument. Added descriptive error messages. * Reverted parameters to old state * Slight changes to the strings printed by logging * Resolved formatting comments * Updated README.md with comand to change logging level * Converted the log level input arguments to lowercase strings * Updated requirements and Readme * Replaced argument version with argument encoding Removed dependency on Results.csv * Readme formatting change * Changed the help strings in argument parser, renamed rnn as fastgrnn * Some more changes to SeeDot-dev.py for the argument parser. Added line that shows that the available fastgrnn model belongs to usps10 dataset. * Added descriptive comments to python files except ONNX and TF folders. * Removed unnecessary comma in launch.json * Corrected default SeeDot values in README.md * Minor changes to improv README.md's readablility. * Added legacy_scales.csv and restored the legacy_scales processor in SeeDot-dev.py * Replaced rnn keyword with fastgrnn in launch.json * Changed all comments to a single format * Added face detection run instructions. Removed lsf flag from README and uppressed it in help. * Replaced maximisingMetric with metric in the main.py, config.py and test.py. Made the respective changes in README.md. * Removed instructions to run face detection * Architecture.md started * Architecture.md iteration 1 completed * Improved line spacing * Removed incorrect line * Replaced architecture.svg * replaced version with encoding * made FastGRNN on usps10 with reduced disagreemts the default for seedot * Points added to architecture.md * Included suggestions from reviewer. * Added rnnpool.sd to folder of seedot models * Added faceDetection code fetcher * Added face detection run instructions and licenses * Updated face Detection instructions and readme.md * Minor correction to readmes * Added rnnpool as an option in SeeDot-dev.py * Added support for face-2 dataset * Error in cleanup in fetchFDDataset.py * Added information about input to Predictor in faceDetection.md * Included reviewr comments to fix typos * Fixed typos in faceDetection.md * Added code to run the quantized face detection on images * Added instructions to run face detection on extra images in faceDetection.md * Added argument support to scale_image.pu * Fixed typos in faceDetection.md * Added info about the output of eval_fromquant in faceDetection.md * Added instructions to fetch an image in faceDetection.md * Added load_sf condition to printing PASS or FAIL * Added instructions to install libraries for detecting bounding boxes * Minor changes to faceDetection.md and main.cpp * Removed duplicate files from the repository and added instructions to copy them for running image to text converters and vice-versa * Minor change to faceDetection.md * Changes to face detection readme (#4) * Replaced the print stagement using logging (error, warning, info and debug levels used). Added progress bar using tqdm package in performSearch function for the 4 stages of exploration. Changes Antlr Check version to 4.9.1. Added requirements.txt * Added more information to the print statment that prints the stage of exploration. * Added log level as a command line argument. Added descriptive error messages. * Reverted parameters to old state * Slight changes to the strings printed by logging * Resolved formatting comments * Updated README.md with comand to change logging level * Converted the log level input arguments to lowercase strings * Updated requirements and Readme * Replaced argument version with argument encoding Removed dependency on Results.csv * Readme formatting change * Changed the help strings in argument parser, renamed rnn as fastgrnn * Some more changes to SeeDot-dev.py for the argument parser. Added line that shows that the available fastgrnn model belongs to usps10 dataset. * Added descriptive comments to python files except ONNX and TF folders. * Removed unnecessary comma in launch.json * Corrected default SeeDot values in README.md * Minor changes to improv README.md's readablility. * Added legacy_scales.csv and restored the legacy_scales processor in SeeDot-dev.py * Replaced rnn keyword with fastgrnn in launch.json * Changed all comments to a single format * Added face detection run instructions. Removed lsf flag from README and uppressed it in help. * Replaced maximisingMetric with metric in the main.py, config.py and test.py. Made the respective changes in README.md. * Removed instructions to run face detection * Architecture.md started * Architecture.md iteration 1 completed * Improved line spacing * Removed incorrect line * Replaced architecture.svg * replaced version with encoding * made FastGRNN on usps10 with reduced disagreemts the default for seedot * Points added to architecture.md * Included suggestions from reviewer. * Added rnnpool.sd to folder of seedot models * Added faceDetection code fetcher * Added face detection run instructions and licenses * Updated face Detection instructions and readme.md * Minor correction to readmes * Added rnnpool as an option in SeeDot-dev.py * Added support for face-2 dataset * Error in cleanup in fetchFDDataset.py * Added information about input to Predictor in faceDetection.md * Included reviewr comments to fix typos * Fixed typos in faceDetection.md * Added code to run the quantized face detection on images * Added instructions to run face detection on extra images in faceDetection.md * Added argument support to scale_image.pu * Fixed typos in faceDetection.md * Added info about the output of eval_fromquant in faceDetection.md * Added instructions to fetch an image in faceDetection.md * Added load_sf condition to printing PASS or FAIL * Added instructions to install libraries for detecting bounding boxes * Minor changes to faceDetection.md and main.cpp * Removed duplicate files from the repository and added instructions to copy them for running image to text converters and vice-versa * Minor change to faceDetection.md * Updated faceDetection.md * Added trace to seedot converstion scripts * Fixed trace dimensions in convert_to_seedot files * Removed torch from the requirements.txt file * Updated face detection readme * Updated faceDetection.md with corrected instructions to install edgeml_pytorch * Reset the changes made to requiments-cpu.txt and requirements-gpu.txt in EdgeML/pytorch * Added multi-line commands to face detection.md * Added instructions for SeeDot installation as well * Some more refactoring * Merged the two convert_to_seedot_scripts into one * fixed typo in convert_RPool_Face_to_SeeDot * Minor addition to faceDetection.md * Minor Nitpicks * Fixed copy instructions in faceDetection.md (#5) * Replaced the print stagement using logging (error, warning, info and debug levels used). Added progress bar using tqdm package in performSearch function for the 4 stages of exploration. Changes Antlr Check version to 4.9.1. Added requirements.txt * Added more information to the print statment that prints the stage of exploration. * Added log level as a command line argument. Added descriptive error messages. * Reverted parameters to old state * Slight changes to the strings printed by logging * Resolved formatting comments * Updated README.md with comand to change logging level * Converted the log level input arguments to lowercase strings * Updated requirements and Readme * Replaced argument version with argument encoding Removed dependency on Results.csv * Readme formatting change * Changed the help strings in argument parser, renamed rnn as fastgrnn * Some more changes to SeeDot-dev.py for the argument parser. Added line that shows that the available fastgrnn model belongs to usps10 dataset. * Added descriptive comments to python files except ONNX and TF folders. * Removed unnecessary comma in launch.json * Corrected default SeeDot values in README.md * Minor changes to improv README.md's readablility. * Added legacy_scales.csv and restored the legacy_scales processor in SeeDot-dev.py * Replaced rnn keyword with fastgrnn in launch.json * Changed all comments to a single format * Added face detection run instructions. Removed lsf flag from README and uppressed it in help. * Replaced maximisingMetric with metric in the main.py, config.py and test.py. Made the respective changes in README.md. * Removed instructions to run face detection * Architecture.md started * Architecture.md iteration 1 completed * Improved line spacing * Removed incorrect line * Replaced architecture.svg * replaced version with encoding * made FastGRNN on usps10 with reduced disagreemts the default for seedot * Points added to architecture.md * Included suggestions from reviewer. * Added rnnpool.sd to folder of seedot models * Added faceDetection code fetcher * Added face detection run instructions and licenses * Updated face Detection instructions and readme.md * Minor correction to readmes * Added rnnpool as an option in SeeDot-dev.py * Added support for face-2 dataset * Error in cleanup in fetchFDDataset.py * Added information about input to Predictor in faceDetection.md * Included reviewr comments to fix typos * Fixed typos in faceDetection.md * Added code to run the quantized face detection on images * Added instructions to run face detection on extra images in faceDetection.md * Added argument support to scale_image.pu * Fixed typos in faceDetection.md * Added info about the output of eval_fromquant in faceDetection.md * Added instructions to fetch an image in faceDetection.md * Added load_sf condition to printing PASS or FAIL * Added instructions to install libraries for detecting bounding boxes * Minor changes to faceDetection.md and main.cpp * Removed duplicate files from the repository and added instructions to copy them for running image to text converters and vice-versa * Minor change to faceDetection.md * Updated faceDetection.md * Added trace to seedot converstion scripts * Fixed trace dimensions in convert_to_seedot files * Removed torch from the requirements.txt file * Updated face detection readme * Updated faceDetection.md with corrected instructions to install edgeml_pytorch * Reset the changes made to requiments-cpu.txt and requirements-gpu.txt in EdgeML/pytorch * Added multi-line commands to face detection.md * Added instructions for SeeDot installation as well * Some more refactoring * Merged the two convert_to_seedot_scripts into one * fixed typo in convert_RPool_Face_to_SeeDot * Minor addition to faceDetection.md * Fixed location of copying from SCUT_Head_Part_B * Minor addition to faceDetection.md * GCC version check added to build.py (#6) * Replaced the print stagement using logging (error, warning, info and debug levels used). Added progress bar using tqdm package in performSearch function for the 4 stages of exploration. Changes Antlr Check version to 4.9.1. Added requirements.txt * Added more information to the print statment that prints the stage of exploration. * Added log level as a command line argument. Added descriptive error messages. * Reverted parameters to old state * Slight changes to the strings printed by logging * Resolved formatting comments * Updated README.md with comand to change logging level * Converted the log level input arguments to lowercase strings * Updated requirements and Readme * Replaced argument version with argument encoding Removed dependency on Results.csv * Readme formatting change * Changed the help strings in argument parser, renamed rnn as fastgrnn * Some more changes to SeeDot-dev.py for the argument parser. Added line that shows that the available fastgrnn model belongs to usps10 dataset. * Added descriptive comments to python files except ONNX and TF folders. * Removed unnecessary comma in launch.json * Corrected default SeeDot values in README.md * Minor changes to improv README.md's readablility. * Added legacy_scales.csv and restored the legacy_scales processor in SeeDot-dev.py * Replaced rnn keyword with fastgrnn in launch.json * Changed all comments to a single format * Added face detection run instructions. Removed lsf flag from README and uppressed it in help. * Replaced maximisingMetric with metric in the main.py, config.py and test.py. Made the respective changes in README.md. * Removed instructions to run face detection * Architecture.md started * Architecture.md iteration 1 completed * Improved line spacing * Removed incorrect line * Replaced architecture.svg * replaced version with encoding * made FastGRNN on usps10 with reduced disagreemts the default for seedot * Points added to architecture.md * Included suggestions from reviewer. * Added rnnpool.sd to folder of seedot models * Added faceDetection code fetcher * Added face detection run instructions and licenses * Updated face Detection instructions and readme.md * Minor correction to readmes * Added rnnpool as an option in SeeDot-dev.py * Added support for face-2 dataset * Error in cleanup in fetchFDDataset.py * Added information about input to Predictor in faceDetection.md * Included reviewr comments to fix typos * Fixed typos in faceDetection.md * Added code to run the quantized face detection on images * Added instructions to run face detection on extra images in faceDetection.md * Added argument support to scale_image.pu * Fixed typos in faceDetection.md * Added info about the output of eval_fromquant in faceDetection.md * Added instructions to fetch an image in faceDetection.md * Added load_sf condition to printing PASS or FAIL * Added instructions to install libraries for detecting bounding boxes * Minor changes to faceDetection.md and main.cpp * Removed duplicate files from the repository and added instructions to copy them for running image to text converters and vice-versa * Minor change to faceDetection.md * Updated faceDetection.md * Added trace to seedot converstion scripts * Fixed trace dimensions in convert_to_seedot files * Removed torch from the requirements.txt file * Updated face detection readme * Updated faceDetection.md with corrected instructions to install edgeml_pytorch * Reset the changes made to requiments-cpu.txt and requirements-gpu.txt in EdgeML/pytorch * Added multi-line commands to face detection.md * Added instructions for SeeDot installation as well * Some more refactoring * Merged the two convert_to_seedot_scripts into one * fixed typo in convert_RPool_Face_to_SeeDot * Minor addition to faceDetection.md * Fixed location of copying from SCUT_Head_Part_B * Minor addition to faceDetection.md * Gcc version restriction added to seedot/Predictor/Makefile * Added GCC version check to SeeDot * Removed extra new-line * Data processing (#7) * Replaced the print stagement using logging (error, warning, info and debug levels used). Added progress bar using tqdm package in performSearch function for the 4 stages of exploration. Changes Antlr Check version to 4.9.1. Added requirements.txt * Added more information to the print statment that prints the stage of exploration. * Added log level as a command line argument. Added descriptive error messages. * Reverted parameters to old state * Slight changes to the strings printed by logging * Resolved formatting comments * Updated README.md with comand to change logging level * Converted the log level input arguments to lowercase strings * Updated requirements and Readme * Replaced argument version with argument encoding Removed dependency on Results.csv * Readme formatting change * Changed the help strings in argument parser, renamed rnn as fastgrnn * Some more changes to SeeDot-dev.py for the argument parser. Added line that shows that the available fastgrnn model belongs to usps10 dataset. * Added descriptive comments to python files except ONNX and TF folders. * Removed unnecessary comma in launch.json * Corrected default SeeDot values in README.md * Minor changes to improv README.md's readablility. * Added legacy_scales.csv and restored the legacy_scales processor in SeeDot-dev.py * Replaced rnn keyword with fastgrnn in launch.json * Changed all comments to a single format * Added face detection run instructions. Removed lsf flag from README and uppressed it in help. * Replaced maximisingMetric with metric in the main.py, config.py and test.py. Made the respective changes in README.md. * Removed instructions to run face detection * Architecture.md started * Architecture.md iteration 1 completed * Improved line spacing * Removed incorrect line * Replaced architecture.svg * replaced version with encoding * made FastGRNN on usps10 with reduced disagreemts the default for seedot * Points added to architecture.md * Included suggestions from reviewer. * Added rnnpool.sd to folder of seedot models * Added faceDetection code fetcher * Added face detection run instructions and licenses * Updated face Detection instructions and readme.md * Minor correction to readmes * Added rnnpool as an option in SeeDot-dev.py * Added support for face-2 dataset * Error in cleanup in fetchFDDataset.py * Added information about input to Predictor in faceDetection.md * Included reviewr comments to fix typos * Fixed typos in faceDetection.md * Added code to run the quantized face detection on images * Added instructions to run face detection on extra images in faceDetection.md * Added argument support to scale_image.pu * Fixed typos in faceDetection.md * Added info about the output of eval_fromquant in faceDetection.md * Added instructions to fetch an image in faceDetection.md * Added load_sf condition to printing PASS or FAIL * Added instructions to install libraries for detecting bounding boxes * Minor changes to faceDetection.md and main.cpp * Removed duplicate files from the repository and added instructions to copy them for running image to text converters and vice-versa * Minor change to faceDetection.md * Updated faceDetection.md * Added trace to seedot converstion scripts * Fixed trace dimensions in convert_to_seedot files * Removed torch from the requirements.txt file * Updated face detection readme * Updated faceDetection.md with corrected instructions to install edgeml_pytorch * Reset the changes made to requiments-cpu.txt and requirements-gpu.txt in EdgeML/pytorch * Added multi-line commands to face detection.md * Added instructions for SeeDot installation as well * Some more refactoring * Merged the two convert_to_seedot_scripts into one * fixed typo in convert_RPool_Face_to_SeeDot * Minor addition to faceDetection.md * Fixed location of copying from SCUT_Head_Part_B * Minor addition to faceDetection.md * Gcc version restriction added to seedot/Predictor/Makefile * Added GCC version check to SeeDot * Removed extra new-line * Added data preprocessing line * The script to fix ranges (#8) * Replaced the print stagement using logging (error, warning, info and debug levels used). Added progress bar using tqdm package in performSearch function for the 4 stages of exploration. Changes Antlr Check version to 4.9.1. Added requirements.txt * Added more information to the print statment that prints the stage of exploration. * Added log level as a command line argument. Added descriptive error messages. * Reverted parameters to old state * Slight changes to the strings printed by logging * Resolved formatting comments * Updated README.md with comand to change logging level * Converted the log level input arguments to lowercase strings * Updated requirements and Readme * Replaced argument version with argument encoding Removed dependency on Results.csv * Readme formatting change * Changed the help strings in argument parser, renamed rnn as fastgrnn * Some more changes to SeeDot-dev.py for the argument parser. Added line that shows that the available fastgrnn model belongs to usps10 dataset. * Added descriptive comments to python files except ONNX and TF folders. * Removed unnecessary comma in launch.json * Corrected default SeeDot values in README.md * Minor changes to improv README.md's readablility. * Added legacy_scales.csv and restored the legacy_scales processor in SeeDot-dev.py * Replaced rnn keyword with fastgrnn in launch.json * Changed all comments to a single format * Added face detection run instructions. Removed lsf flag from README and uppressed it in help. * Replaced maximisingMetric with metric in the main.py, config.py and test.py. Made the respective changes in README.md. * Removed instructions to run face detection * Architecture.md started * Architecture.md iteration 1 completed * Improved line spacing * Removed incorrect line * Replaced architecture.svg * replaced version with encoding * made FastGRNN on usps10 with reduced disagreemts the default for seedot * Points added to architecture.md * Included suggestions from reviewer. * Added rnnpool.sd to folder of seedot models * Added faceDetection code fetcher * Added face detection run instructions and licenses * Updated face Detection instructions and readme.md * Minor correction to readmes * Added rnnpool as an option in SeeDot-dev.py * Added support for face-2 dataset * Error in cleanup in fetchFDDataset.py * Added information about input to Predictor in faceDetection.md * Included reviewr comments to fix typos * Fixed typos in faceDetection.md * Added code to run the quantized face detection on images * Added instructions to run face detection on extra images in faceDetection.md * Added argument support to scale_image.pu * Fixed typos in faceDetection.md * Added info about the output of eval_fromquant in faceDetection.md * Added instructions to fetch an image in faceDetection.md * Added load_sf condition to printing PASS or FAIL * Added instructions to install libraries for detecting bounding boxes * Minor changes to faceDetection.md and main.cpp * Removed duplicate files from the repository and added instructions to copy them for running image to text converters and vice-versa * Minor change to faceDetection.md * Updated faceDetection.md * Added trace to seedot converstion scripts * Fixed trace dimensions in convert_to_seedot files * Removed torch from the requirements.txt file * Updated face detection readme * Updated faceDetection.md with corrected instructions to install edgeml_pytorch * Reset the changes made to requiments-cpu.txt and requirements-gpu.txt in EdgeML/pytorch * Added multi-line commands to face detection.md * Added instructions for SeeDot installation as well * Some more refactoring * Merged the two convert_to_seedot_scripts into one * fixed typo in convert_RPool_Face_to_SeeDot * Minor addition to faceDetection.md * Fixed location of copying from SCUT_Head_Part_B * Minor addition to faceDetection.md * Gcc version restriction added to seedot/Predictor/Makefile * Added GCC version check to SeeDot * Removed extra new-line * Added data preprocessing line * Added cpu conversion in convert_RPool_Face_to_SeeDot.py * Added script to process seedot input files * Increased precision of printing floating point in fixSeeDotInput.py * Updated instructions to update ranges by script * Add newline at the end of fixSeeDotInput.py * ONNX dependency added to README (#9) * Replaced the print stagement using logging (error, warning, info and debug levels used). Added progress bar using tqdm package in performSearch function for the 4 stages of exploration. Changes Antlr Check version to 4.9.1. Added requirements.txt * Added more information to the print statment that prints the stage of exploration. * Added log level as a command line argument. Added descriptive error messages. * Reverted parameters to old state * Slight changes to the strings printed by logging * Resolved formatting comments * Updated README.md with comand to change logging level * Converted the log level input arguments to lowercase strings * Updated requirements and Readme * Replaced argument version with argument encoding Removed dependency on Results.csv * Readme formatting change * Changed the help strings in argument parser, renamed rnn as fastgrnn * Some more changes to SeeDot-dev.py for the argument parser. Added line that shows that the available fastgrnn model belongs to usps10 dataset. * Added descriptive comments to python files except ONNX and TF folders. * Removed unnecessary comma in launch.json * Corrected default SeeDot values in README.md * Minor changes to improv README.md's readablility. * Added legacy_scales.csv and restored the legacy_scales processor in SeeDot-dev.py * Replaced rnn keyword with fastgrnn in launch.json * Changed all comments to a single format * Added face detection run instructions. Removed lsf flag from README and uppressed it in help. * Replaced maximisingMetric with metric in the main.py, config.py and test.py. Made the respective changes in README.md. * Removed instructions to run face detection * Architecture.md started * Architecture.md iteration 1 completed * Improved line spacing * Removed incorrect line * Replaced architecture.svg * replaced version with encoding * made FastGRNN on usps10 with reduced disagreemts the default for seedot * Points added to architecture.md * Included suggestions from reviewer. * Added rnnpool.sd to folder of seedot models * Added faceDetection code fetcher * Added face detection run instructions and licenses * Updated face Detection instructions and readme.md * Minor correction to readmes * Added rnnpool as an option in SeeDot-dev.py * Added support for face-2 dataset * Error in cleanup in fetchFDDataset.py * Added information about input to Predictor in faceDetection.md * Included reviewr comments to fix typos * Fixed typos in faceDetection.md * Added code to run the quantized face detection on images * Added instructions to run face detection on extra images in faceDetection.md * Added argument support to scale_image.pu * Fixed typos in faceDetection.md * Added info about the output of eval_fromquant in faceDetection.md * Added instructions to fetch an image in faceDetection.md * Added load_sf condition to printing PASS or FAIL * Added instructions to install libraries for detecting bounding boxes * Minor changes to faceDetection.md and main.cpp * Removed duplicate files from the repository and added instructions to copy them for running image to text converters and vice-versa * Minor change to faceDetection.md * Updated faceDetection.md * Added trace to seedot converstion scripts * Fixed trace dimensions in convert_to_seedot files * Removed torch from the requirements.txt file * Updated face detection readme * Updated faceDetection.md with corrected instructions to install edgeml_pytorch * Reset the changes made to requiments-cpu.txt and requirements-gpu.txt in EdgeML/pytorch * Added multi-line commands to face detection.md * Added instructions for SeeDot installation as well * Some more refactoring * Merged the two convert_to_seedot_scripts into one * fixed typo in convert_RPool_Face_to_SeeDot * Minor addition to faceDetection.md * Fixed location of copying from SCUT_Head_Part_B * Minor addition to faceDetection.md * Gcc version restriction added to seedot/Predictor/Makefile * Added GCC version check to SeeDot * Removed extra new-line * Added data preprocessing line * Added cpu conversion in convert_RPool_Face_to_SeeDot.py * Added script to process seedot input files * Increased precision of printing floating point in fixSeeDotInput.py * Updated instructions to update ranges by script * Add newline at the end of fixSeeDotInput.py * Added ONNX as a dependencey in README * Minor Fixes * Improve README Co-authored-by: ShikharJ <jaiswalshikhar87@gmail.com> Co-authored-by: G Rahul Kranti Kiran <krantikiran.68@gmail.com> |
||
|---|---|---|
| .. | ||
| include | ||
| models | ||
| src | ||
| tests | ||
| Makefile | ||
| README.md | ||
| config.mk | ||
README.md
Copyright (c) Microsoft Corporation. All rights reserved.
Licensed under the MIT license.
This folder consists of C reference code for some EdgeML operators,
and sample parameter sets and input/output traces to verify them.
The code is intended to compile with GCC on Linux machines,
and is to be adapted as needed for other embedded platforms.
Directory Structure
The EdgeML/c_reference/ directory is broadly structured into the following sub-directories:
- include/: Contains the header files for various lower level operators and layers.
- models/: Contains the optimized source code and header files for various models built by stiching together different layers and operators. Also contains the layer weights and hyper-parameters for the corresponding models as well (stored using
Git LFS). (Note: Cloning the repo without installingGit LFSwould fail to clone the actual headers. It's recommended to follow instructions on setting upLFSfrom here before cloning.) - src/: Contains the optimized source code files for various lower level operators and layers.
- tests/: Contains extensive test cases for individual operators and layers, as well as the implemented models. The executables are generated in the main directory itself, while the test scripts and their configurations can be accessed in the appropriate sub-directories.
Compiling
Run make inside the EdgeML/c_reference/ directory to compile the entire project at once. Alternatively, run make clean to discard the previously generated object files and executables. By default, the directory is compiled with loop unrolling and shift operations turned off.
Running
Head to c_reference/tests/ directory and execute the test script of your choice. Test patches (wherever required) are currently not included because of license restrictions. Please open an issue / refer to an existing issue for the same.