• Automation framework: CK
  
• Development repository: ck-ml
  
• Source: GitHub
  
• Available command lines:
  
  • ck run program:image-classification-resnet-50-cntk --cmd_key=train (META)
    
• Support for host OS: any
  
• Support for target OS: any
  
• Tags: image-classification,cntk-image-classification
  
• How to get the stable version via the client:
  
  pip install cbench
  cb download program:image-classification-resnet-50-cntk --version=3.0.0 --all
  ck run program:image-classification-resnet-50-cntk
• How to get the development version:
  
  pip install ck
  ck pull repo:ck-ml
  ck run program:image-classification-resnet-50-cntk
  

ImageClassification-ResNet-CNTK

This folder contains the CNTK python implementation of the ResNet model. The original official repo is here.

Prepare Dataset

The original directory of CNTK ResNet contains code for cifar10 and imagenet datasets. In our benchmark, we use the imagenet1K dataset.

To use our benchmark, first prepare the dataset according to the following steps:

1, Download and decompress the imagenet1K 2012 dataset. Note that you need to sign up for an account at image-net.org to download the dataset. Rename and move your deceompressed directories so that the raw data folder looks like this:

ImageNet1K/
    ImageNet_train/
        n01440764/
        n01443537/
        n01484850/
        ...
    ImageNet_val/
        ILSVRC2012_val_00000001.JPEG
        ILSVRC2012_val_00000002.JPEG
        ILSVRC2012_val_00000003.JPEG
        ...

2, Copy dataset/Read_labelclsloc.py & dataset/create_train_map.py to ImageNet1K/ImageNet_train/, and dataset/ILSVRC2012_validation_ground_truth.txt & dataset/create_val_map.py to ImageNet1K/ImageNet_val/

3, Generate train_map.txt and val_map.txt by:

DATA_DIR=<ABS_PATH_TO_ImageNet1K/ImageNet_train/>
cd $DATA_DIR
python3.5 create_train_map.py $DATA_DIR
cd ../ImageNet_val
python create_val_map.py

Get CNTK Executable

Currently we provide CNTK BrainScript supported by CNTK-2.0. We will update to latest version of CNTK in future release.

To train the model with CNTK BrainScript, obtain the CNTK executable by following:\ 1, Install open MPI by sudo apt-get install libopenmpi-dev;\ 2, Download and decompress CNTK-2.0 GPU version for Linux here. To use other version, check here (Please download the release, not the source code);\ 3, Add dependency libs to the environment by:

export LD_LIBRARY_PATH=LD_LIBRARY_PATH:<PATH/TO/cntk/cntk/dependencies/lib>

4, You should find the cntk executable in <PATH/TO/cntk/cntk/bin/cntk>.

Run Training

To start the training of ResNet-50 with mini-batch size x with CNTK BrainScript, you need to:\ 1, Change the DataDir variable in source/BrainScript/ResNet50_ImageNet1K.cntk to <ABS_PATH_TO_ImageNet1K/ImageNet_train/>;\ 2, Modify the minibatchSize variable to x;\ 3, start the training by:

<PATH/TO/cntk/cntk/bin/cntk> configFile=ResNet50_ImageNet1K.cntk

To train ResNet-152, change the DataDir & minibatchSize variables accordingly, then start the training by:

<PATH/TO/cntk/cntk/bin/cntk> configFile=ResNet152_ImageNet1K.cntk

• 3.0.0 (2021-04-12)
• 1.0.0 (2019-10-08)

• .cm/desc.json (3 bytes)
• .cm/info.json (320 bytes)
• .cm/meta.json (1742 bytes)
• README.md (2762 bytes)
• source/BrainScript/Macros.bs (3542 bytes)
• source/BrainScript/ResNet101_ImageNet1K.cntk (5590 bytes)
• source/BrainScript/ResNet152_ImageNet1K.cntk (5608 bytes)
• source/BrainScript/ResNet18_ImageNet1K.cntk (5246 bytes)
• source/BrainScript/ResNet34_ImageNet1K.cntk (5324 bytes)
• source/BrainScript/ResNet50_ImageNet1K.cntk (5100 bytes)
• source/Python/TrainResNet_ImageNet_Distributed.py (11239 bytes)
• source/Python/resnet_models.py (5417 bytes)
• source/requirements.gpu-cu80.txt (1 bytes)