Home

Awesome

VoxelPose

This is the official implementation for:

VoxelPose: Towards Multi-Camera 3D Human Pose Estimation in Wild Environment,
Hanyue Tu, Chunyu Wang, Wenjun Zeng
ECCV 2020 (Oral) (arXiv 2004.06239)

<img src="data/panoptic2.gif" width="800"/>

Installation

  1. Clone this repo, and we'll call the directory that you cloned multiview-multiperson-pose as ${POSE_ROOT}.
  2. Install dependencies.

Data preparation

Shelf/Campus datasets

  1. Download the datasets from http://campar.in.tum.de/Chair/MultiHumanPose and extract them under ${POSE_ROOT}/data/Shelf and ${POSE_ROOT}/data/CampusSeq1, respectively.

  2. We have processed the camera parameters to our formats and you can download them from this repository. They lie in ${POSE_ROOT}/data/Shelf/ and ${POSE_ROOT}/data/CampusSeq1/, respectively.

  3. Due to the limited and incomplete annotations of the two datasets, we don't train our model using this dataset. Instead, we directly use the 2D pose estimator trained on COCO, and use independent 3D human poses from the Panoptic dataset to train our 3D model. It lies in ${POSE_ROOT}/data/panoptic_training_pose.pkl. See our paper for more details.

  4. For testing, we first estimate 2D poses and generate 2D heatmaps for these two datasets in this repository. The predicted poses can also download from the repository. They lie in ${POSE_ROOT}/data/Shelf/ and ${POSE_ROOT}/data/CampusSeq1/, respectively. You can also use the models trained on COCO dataset (like HigherHRNet) to generate 2D heatmaps directly.

The directory tree should look like this:

${POSE_ROOT}
|-- data
    |-- Shelf
    |   |-- Camera0
    |   |-- ...
    |   |-- Camera4
    |   |-- actorsGT.mat
    |   |-- calibration_shelf.json
    |   |-- pred_shelf_maskrcnn_hrnet_coco.pkl
    |-- CampusSeq1
    |   |-- Camera0
    |   |-- Camera1
    |   |-- Camera2
    |   |-- actorsGT.mat
    |   |-- calibration_campus.json
    |   |-- pred_campus_maskrcnn_hrnet_coco.pkl
    |-- panoptic_training_pose.pkl

CMU Panoptic dataset

  1. Download the dataset by following the instructions in panoptic-toolbox and extract them under ${POSE_ROOT}/data/panoptic_toolbox/data.
  1. Download the pretrained backbone model from pretrained backbone and place it here: ${POSE_ROOT}/models/pose_resnet50_panoptic.pth.tar (ResNet-50 pretrained on COCO dataset and finetuned jointly on Panoptic dataset and MPII).

The directory tree should look like this:

${POSE_ROOT}
|-- models
|   |-- pose_resnet50_panoptic.pth.tar
|-- data
    |-- panoptic-toolbox
        |-- data
            |-- 16060224_haggling1
            |   |-- hdImgs
            |   |-- hdvideos
            |   |-- hdPose3d_stage1_coco19
            |   |-- calibration_160224_haggling1.json
            |-- 160226_haggling1  
            |-- ...

Training

CMU Panoptic dataset

Train and validate on the five selected camera views. You can specify the GPU devices and batch size per GPU in the config file. We trained our models on two GPUs.

python run/train_3d.py --cfg configs/panoptic/resnet50/prn64_cpn80x80x20_960x512_cam5.yaml

Shelf/Campus datasets

python run/train_3d.py --cfg configs/shelf/prn64_cpn80x80x20.yaml
python run/train_3d.py --cfg configs/campus/prn64_cpn80x80x20.yaml

Evaluation

CMU Panoptic dataset

Evaluate the models. It will print evaluation results to the screen./

python test/evaluate.py --cfg configs/panoptic/resnet50/prn64_cpn80x80x20_960x512_cam5.yaml

Shelf/Campus datasets

It will print the PCP results to the screen.

python test/evaluate.py --cfg configs/shelf/prn64_cpn80x80x20.yaml
python test/evaluate.py --cfg configs/campus/prn64_cpn80x80x20.yaml

Citation

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

@inproceedings{voxelpose,
    author={Tu, Hanyue and Wang, Chunyu and Zeng, Wenjun},
    title={VoxelPose: Towards Multi-Camera 3D Human Pose Estimation in Wild Environment},
    booktitle = {European Conference on Computer Vision (ECCV)},
    year = {2020}
}

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.