Home

Awesome

HumanGPS: Geodesic PreServing Feature for Dense Human Correspondences

Tensorflow implementation of the paper "HumanGPS: Geodesic PreServing Feature for Dense Human Correspondences", CVPR 2021.

Project Page | Videos | Paper | Data

Setup

pip install -r requirements.txt  --user
pip install gdown

Running code

Here we show how to run our code on SMPL intra and inter testing data. You can download the rest of the synthetic SMPL testing data used in the paper here.

1. Download pretrained model.

bash download_model.sh

2. Evaluate on intra testing data.

(a) Run

mkdir -p ./test_data/

Download our SMPL intra test data from smpl_intra_data in ./test_data/

To evaluate average epe on intra test dataset.

(b) set JOB_NAME="eval_optical_flow_intra" in ./script/inference_local.sh

(c) Run

bash ./script/inference_local.sh

3. Evaluate on inter testing data.

(a) Run

mkdir -p ./test_data/

Download our SMPL inter test data from smpl_inter_data in ./test_data/

To evaluate average epe on inter test dataset.

(b) set JOB_NAME="eval_optical_flow_inter" in ./script/inference_local.sh

(c) Run

bash ./script/inference_local.sh

4. Train on intra testing data.

Currently, we can not provide the whole training dataset due to the copyright and huge size of the data.

Here, we provide an example configuration for training on intra testing data.

bash ./script/train_local.sh

5. Inference on toy examples for visualization.

Check out ./inference_demo.ipynb for toy examples.

Citation

If you find this code useful in your research, please cite:

@inproceedings{tan2021humangps,
  title = {{HumanGPS: Geodesic PreServing Feature for Dense Human Correspondence}},
  author = {Tan, Feitong and Tang, Danhang and Dou, Mingsong and Guo, Kaiwen and Pandey, Rohit and Keskin, Cem and Du, Ruofei and Sun, Deqing and Bouaziz, Sofien and Fanello, Sean and Tan, Ping and Zhang, Yinda},
  booktitle = {2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2021},
  publisher = {IEEE},
}