This is an official pytorch implementation of MEBOW: Monocular Estimation of Body Orientation In the Wild. In this work, we present COCO-MEBOW (Monocular Estimation of Body Orientation in the Wild), a new large-scale dataset for orientation estimation from a single in-the-wild image. Based on COCO-MEBOW, we established a simple baseline model for human body orientation estimation. This repo provides the code.

1. Install pytorch >= v1.0.0 following official instruction.

2. Clone this repo, and we'll call the directory that you cloned as ${HBOE_ROOT}.

3. Install dependencies:

   pip install -r requirements.txt
   

4. Install COCOAPI:

   # COCOAPI=/path/to/clone/cocoapi
   git clone https://github.com/cocodataset/cocoapi.git $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
   

   Note that instructions like # COCOAPI=/path/to/install/cocoapi indicate that you should pick a path where you'd like to have the software cloned and then set an environment variable (COCOAPI in this case) accordingly.

5. Init output(training model output directory) and log(tensorboard log directory) directory:

   mkdir output 
   mkdir log
   

   Your directory tree should look like this:

   ${HBOE_ROOT}
   ├── data
   ├── experiments
   ├── lib
   ├── log
   ├── models
   ├── output
   ├── tools 
   ├── README.md
   └── requirements.txt
   

6. Download pretrained models from the model zoo provided by HRnet(GoogleDrive or OneDrive)

   ${HBOE_ROOT}
    `-- models
        `-- pose_hrnet_w32_256x192.pth
   

For MEBOW dataset, please download images, bbox and keypoints annotation from COCO download. Please email [email protected] to get access to human body orientation annotation. Put them under {HBOE_ROOT}/data, and make them look like this:

${HBOE_ROOT}
|-- data
`-- |-- coco
    `-- |-- annotations
        |   |-- train_hoe.json
        |   |-- val_hoe.json
        |   |-- person_keypoints_train2017.json
        |   `-- person_keypoints_val2017.json
        `-- images
            |-- train2017
            |   |-- 000000000009.jpg
            |   |-- 000000000025.jpg
            |   |-- 000000000030.jpg
            |   |-- ... 
            `-- val2017
                |-- 000000000139.jpg
                |-- 000000000285.jpg
                |-- 000000000632.jpg
                |-- ... 

For TUD dataset, please download images from the web page of TUD. The page also provides 8-bin orientation annotation. Continuous orientation annotation for TUD dataset can be found from here. We provide our precessed TUD annotation from here. Put TUD images and our processed annotation under {HBOE_ROOT}/data, and make them look like this:

${HBOE_ROOT}
|-- data
`-- |-- tud
    `-- |-- annot
        |   |-- train_tud.pkl
        |   |-- val_tud.pkl
        |   `-- test_tud.pkl
        `-- images
            |-- train
            |-- validate
            `-- test

We also provide the trained HBOE model (MEBOW as training set). (OneDrive)

python tools/train.py --cfg experiments/coco/segm-4_lr1e-3.yaml

python tools/train.py --cfg experiments/tud/lr1e-3.yaml

You should change TEST:MODEL_FILE to your own in "experiments/coco/segm-4_lr1e-3.yaml". If you want to test with our trained HBOE model, specify TEST:MODEL_FILE with the downloaded model path.

python tools/test.py --cfg experiments/coco/segm-4_lr1e-3.yaml

This repo is based on HRnet.

If you use our dataset or models in your research, please cite with:

@inproceedings{wu2020mebow,
  title={MEBOW: Monocular Estimation of Body Orientation In the Wild},
  author={Wu, Chenyan and Chen, Yukun and Luo, Jiajia and Su, Che-Chun and Dawane, Anuja and Hanzra, Bikramjot and Deng, Zhuo and Liu, Bilan and Wang, James Z and Kuo, Cheng-hao},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={3451--3461},
  year={2020}
}