CVWC2019-pose

Introduction

Using HRNet for CVWC 2019 Tiger Pose Track Challenge based on https://github.com/leoxiaobin/deep-high-resolution-net.pytorch. For the keypoints difference between human and tiger, we made some modifications in the original cocoapi.

Environment

The code is developed using python 3.6.6 on Ubuntu 16.04. NVIDIA GPUs are needed. The code is developed and tested using 1 Tesla V100 GPU cards. Other platforms or GPU cards are not fully tested, and you need pay attention to the modification of the parameters in cfg file, like LR, END_EPOCH, BATCH_SIZE_PER_GPU and etc. You can refer to the scheduling rules from Detectron.

Installation

1. Install pytorch >= v1.0.0 following official instruction. Note that if you use pytorch's version < v1.0.0, you should following the instruction at https://github.com/Microsoft/human-pose-estimation.pytorch to disable cudnn's implementations of BatchNorm layer. We encourage you to use higher pytorch's version(>=v1.0.0)

2. Clone this repo, and we'll call the directory './CVWC2019-pose/deep-high-resolution-net.pytorch' you cloned as ${POSE_ROOT}. And './CVWC2019-pose/cocoapi' as ${COCOAPI}.

3. Install dependencies under {POSE_ROOT}:

   pip install -r requirements.txt
   

4. Make libs:

   cd ${POSE_ROOT}/lib
   make
   

5. Install modified cocoapi:

   cd ${COCOAPI}/PythonAPI
   # Install into global site-packages
   make install
   # Alternatively, if you do not have permissions or prefer
   # not to install the COCO API into global site-packages
   python3 setup.py install --user
   

6. Init output(training model output directory) and log(tensorboard log directory) directory under ${POSE_ROOT}:

   mkdir output 
   mkdir log
   

   Your directory tree should looks like this:

   ${POSE_ROOT}
   ├── data
   ├── experiments
   ├── lib
   ├── log
   ├── models
   ├── output
   ├── tools 
   ├── train.sh
   ├── test.sh
   └── requirements.txt
   

7. Download pretrained models from HRNet model zoo(GoogleDrive or OneDrive)

   ${POSE_ROOT}
    `-- models
        `-- pytorch
            |-- imagenet
            |   |-- hrnet_w48-8ef0771d.pth
   
   

Data Preparation

For CVWC 2019 Tiger Pose Track data, you should make them like this:

${POSE_ROOT}
|-- data
`-- |-- atrw
    `-- |-- pose
        `-- |-- annotations
            |   |-- image_info_test.json
            |   |-- keypoints_train.json
            |   |-- keypoints_trainval.json
            |   |-- keypoints_val.json
            `-- images
                `-- test
                |   |-- 000000.jpg
                |   |-- 000004.jpg
                `-- train
                    |-- 000002.jpg
                    |-- 000003.jpg
                    |-- ... 
                `-- val
                    |-- 000001.jpg
                    |-- 000013.jpg
                    |-- ... 
                `-- trainval
                    |-- 000001.jpg
                    |-- 000002.jpg
                    |-- ...

You can directly download the json files from dir 'data/atrw/pose/annotations'.

Training and Testing

Please pay attention to modify the parameter of DATASET in cfg file for training or testing.

Train on tiger dataset using pretrained model:

bash train.sh

Test with our best model:

bash test.sh

Results

Citation

Thanks for MSRA's wonderful work

@inproceedings{sun2019deep,
  title={Deep High-Resolution Representation Learning for Human Pose Estimation},
  author={Sun, Ke and Xiao, Bin and Liu, Dong and Wang, Jingdong},
  booktitle={CVPR},
  year={2019}
}

@inproceedings{xiao2018simple,
    author={Xiao, Bin and Wu, Haiping and Wei, Yichen},
    title={Simple Baselines for Human Pose Estimation and Tracking},
    booktitle = {European Conference on Computer Vision (ECCV)},
    year = {2018}
}