ShiqiYu/OpenGait

A flexible and extensible framework for gait recognition. You can focus on designing your own models and comparing with state-of-the-arts easily with the help of OpenGait.

OpenGait – A Comprehensive Gait‑Recognition Research Platform

What it is – OpenGait is an open‑source toolbox for studying gait recognition: identifying people from the way they walk. It bundles datasets, model implementations, training scripts, and evaluation utilities so researchers can quickly try out state‑of‑the‑art methods or develop new ones.

Key capabilities

  • Broad dataset support – ready‑to‑use loaders for the major gait benchmarks (CASIA‑B, OUMVLP, GREW, Gait3D, GaitLU‑1M, SUSTech1K, etc.).
  • Many SOTA models – implementations of appearance‑based, pose‑based, and multimodal methods such as DeepGaitV2, SwinGait, SkeletonGait++, BigGait, FoundationGait, BiggerGait, LidarGait++, DenoisingGait, MultiGait++ and more.
  • Training infrastructure – built on PyTorch with Distributed Data Parallel (DDP) and Automatic Mixed Precision (AMP) support, plus TensorBoard logging.
  • Model zoo – pre‑trained checkpoints for most models are hosted on Hugging Face, making inference easy.
  • Extensible – clear configuration files and a “how‑to‑create‑your‑model” guide let users plug in custom architectures.

Typical workflow

  1. Prepare data – follow the docs/2.prepare_dataset.md tutorial to download and index a supported dataset.
  2. Configure – edit a YAML config (model, dataset, training hyper‑parameters) as described in docs/3.detailed_config.md.
  3. Train / test – run the provided training script; DDP and AMP are enabled automatically for multi‑GPU runs.
  4. Evaluate – results are logged to TensorBoard and can be compared against the published baselines in the model‑zoo docs.

Why it matters – Gait is a privacy‑preserving biometric; OpenGait supplies the research community with a unified, reproducible platform that tracks the rapid progress from classic CNNs to large‑vision‑model‑based foundations and even LiDAR‑based 3‑D gait. The repository also publishes accompanying surveys and benchmark papers, keeping users up‑to‑date with the latest trends.

Getting started – Clone the repo, install the requirements.txt (PyTorch ≥ 2.0), and follow docs/0.get_started.md. Pre‑trained models can be downloaded from the Hugging Face link shown in the README.

Citation – The authors ask users to cite the CVPR 2023 and TPAMI 2025 OpenGait papers when publishing results.

Related

  • Project
    PetoiCamp/OpenCat-Quadruped-Robot

    An open-source framework for quadruped robots that handles gait coordination and servo control, enabling researchers and makers to build AI-powered robotic pets.

  • Project
    dluvizon/deephar

    A multi-task deep learning framework for real-time 2D and 3D human pose estimation and action recognition.

  • Project
    perfanalytics/pose2sim

    An open-source workflow for markerless 3D kinematics that converts 2D pose estimation from standard cameras into research-grade 3D joint angles and skeletal analysis.

  • Project
    open-compass/opencompass

    A comprehensive evaluation platform for large language and vision-language models that integrates over 100 datasets and multiple inference backends for standardized benchmarking.

  • Project
    shaoshengsong/DeepSORT

    A real-time multi-object tracking system written in C++ that implements ByteTrack, OC-SORT, Deep OC-SORT, and DeepSORT using YOLO and ONNX Runtime.