gaitgraph2's Introduction

GaitGraph2: Towards a Deeper Understanding of Skeleton-based Gait Recognition

Accepted at: 17th IEEE Computer Society Workshop on Biometrics 2022 (CVPRW'22)

Previous version:
GaitGraph: Graph Convolutional Network for Skeleton-Based Gait Recognition (ICIP'21)

Preparation

Clone the repository and install the dependencies from requirements.txt.

Datasets

CASIA-B: Download from here and move casia-b_pose_coco.csv to data
OUMVLP-Pose: Download from here

Pre-trained models

You can download pre-trained models from here.

Running the code

We use PyTorch Lightning CLI for configuration and training.

Train:

# CASIA-B
python3 gaitgraph_casia_b.py fit --config configs/casia_b.yaml 
# OUMVLP-Pose (OpenPose)
python3 gaitgraph_oumvlp.py fit --config configs/oumvlp.yaml
# OUMVLP-Pose (AlphaPose)
python3 gaitgraph_oumvlp.py fit --config configs/oumvlp.yaml --data.keypoints alphapose

Test:

python3 gaitgraph_{casia_b,oumvlp}.py predict --config <path_to_config_file> --ckpt_path <path_to_checkpoint> --model.tta True

Logs and checkpoints will be saved to lighting_logs and can be shown in tensorboard with:

tensorboard --logdir lightning_logs

Visualization of Activations

To run the visualization add run the following command:

python3 gaitgraph_{casia_b,oumvlp}.py predict --config <path_to_config_file> --ckpt_path <path_to_checkpoint> --config configs/draw.yaml --model.tta False

Examples:

Citation

If you use GaitGraph2, please use the following BibTeX entry.

@misc{teepe2022gaitgraph2,
  doi = {10.48550/ARXIV.2204.07855},
  url = {https://arxiv.org/abs/2204.07855},
  author = {Teepe, Torben and Gilg, Johannes and Herzog, Fabian and H\"ormann, Stefan and Rigoll, Gerhard},
  title = {Towards a Deeper Understanding of Skeleton-based Gait Recognition},
  publisher = {arXiv},
  year = {2022}
}

Recommend Projects