HSTL

Hierarchical Spatio-Temporal Representation Learning for Gait Recognition

ICCV 2023 · Hierarchical Gait Representation Learning · Official PyTorch Implementation

---

🔥 News

• Jul. 2023: Our paper "Hierarchical Spatio-Temporal Representation Learning for Gait Recognition" was accepted by ICCV 2023.
• This work marks the first paper from Hebei Agricultural University accepted by ICCV since the university was founded more than 120 years ago, achieving a historic 0-to-1 breakthrough for the university in top-tier computer vision conferences.
• This paper was completed with Hebei Agricultural University as the independent affiliation, without external institutional collaboration.
• Related reports: College News · University News

📝 Introduction

This repository provides the official implementation of our ICCV 2023 paper:

Hierarchical Spatio-Temporal Representation Learning for Gait Recognition
Lei Wang, Bo Liu, Fangfang Liang, Bincheng Wang
IEEE/CVF International Conference on Computer Vision (ICCV), 2023

HSTL learns hierarchical spatio-temporal representations for gait recognition. Instead of modeling gait sequences with only flat or isolated part-level representations, HSTL captures motion patterns from coarse to fine levels and improves the representation ability for challenging gait recognition scenarios.

🧭 Project Navigation

Section	Description
Operating Environments	Hardware and software dependencies
Checkpoints	Pre-trained checkpoints for CASIA-B and OUMVLP
Train and Test	Commands for training and evaluation
Acknowledgement	Codebase acknowledgement
Citation	BibTeX citation

🖥️ Operating Environments

Hardware Environment

Our code runs on a server with:

8 × NVIDIA GeForce RTX 3090 GPUs
Intel(R) Core(TM) i7-9800X CPU @ 3.80GHz

Software Environment

PyTorch = 1.10
torchvision
pyyaml
tensorboard
opencv-python
tqdm

📦 Checkpoints

Dataset	Checkpoint
CASIA-B	Google Drive
OUMVLP	Google Drive

🚀 Train and Test

Train

Train a model with:

CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch \
  --nproc_per_node=2 \
  opengait/main.py \
  --cfgs ./configs/htsl/hstl.yaml \
  --phase train

Arguments:

Argument	Description
python -m torch.distributed.launch	DDP launch instruction
--nproc_per_node	Number of GPUs to use; it must equal the length of CUDA_VISIBLE_DEVICES
--cfgs	Path to the config file
--phase train	Run the training phase
--log_to_file	Save terminal logs to disk if specified

You can also run commands in:

bash train.sh

Test

Evaluate a trained model with:

CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch \
  --nproc_per_node=2 \
  opengait/main.py \
  --cfgs ./configs/htsl/hstl.yaml \
  --phase test

Arguments:

Argument	Description
--phase test	Run the testing phase
--iter	Specify the iteration checkpoint for evaluation

Other arguments are the same as the training phase.

You can also run commands in:

bash test.sh

🙏 Acknowledgement

This codebase is based on OpenGait. We sincerely thank the authors for their excellent work.

📚 Citation

If you find this project useful, please consider citing our paper:

@InProceedings{Wang_2023_ICCV,
    author    = {Wang, Lei and Liu, Bo and Liang, Fangfang and Wang, Bincheng},
    title     = {Hierarchical Spatio-Temporal Representation Learning for Gait Recognition},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {19639-19649}
}

⭐ Star History