AMVUR

This is a PyTorch implementation of our paper: A Probabilistic Attention Model with Occlusion-aware Texture Regression for 3D Hand Reconstruction from a Single RGB Image(CVPR 2023)

Requirements

• Python (>=3.7)
• PyTorch (>=1.7.1)
• torchvision (>=0.8.2)
• cuda (>=11.0)
• PyTorch3D (>=0.3.0)

  Data

• Download the FreiHAND dataset from the website.
• Download the HO3D dataset from the website, and annotation files from [here]. You need to put HO3D dataset to ${REPO_DIR}/data file, and annotation files to ${REPO_DIR}/data/HO3D_v2/annotations/ file.

  Pytorch MANO layer

• Download manopth, and put the file to ${REPO_DIR}/manopth.
• Download MANO_RIGHT.pkl from here, and put the file to ${REPO_DIR}/AMVUR/modeling/data.

  Backbone Download

  Download the cls_hrnet_w64_sgd_lr5e-2_wd1e-4_bs32_x100.yaml and hrnetv2_w64_imagenet_pretrained.pth from HRNet models, and put them to ${REPO_DIR}/models/hrnet.

Experiment

• Supervised Experiment

  Evaluation: Our pre-trained model can be downloaded from here, and put the file to ${REPO_DIR}/pre_trained.

  Run:

  python -m torch.distributed.launch --nproc_per_node=4 \
     experiments/supervised_HO3D_v2.py \
     --config_json ./experiments/config/test.json

  It will generate a prediction file called pred.zip. Afte that, please submit the prediction file to codalab challenge and see the results.

  Training:

  python -m torch.distributed.launch --nproc_per_node=4 \
     experiments/supervised_HO3D_v2.py \
     --config_json ./experiments/config/train.json

• Weakly Supervised Experiment

  Evaluation:

  python -m torch.distributed.launch --nproc_per_node=4 \
     experiments/weakly_supervised_HO3D_v2.py \
     --config_json ./experiments/config/test.json

  Training:

  python -m torch.distributed.launch --nproc_per_node=4 \
     experiments/weakly_supervised_HO3D_v2.py \
     --config_json ./experiments/config/train.json

Citation

@inproceedings{jiang2023probabilistic,
  title={A Probabilistic Attention Model with Occlusion-aware Texture Regression for 3D Hand Reconstruction from a Single RGB Image},
  author={Jiang, Zheheng and Rahmani, Hossein and Black, Sue and Williams, Bryan M},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={758--767},
  year={2023}
}