mmpose - 先进的开源姿态估计工具箱

OpenMMLab website ^HOT OpenMMLab platform ^{TRY IT OUT}

Introduction

English | 简体中文

MMPose is an open-source toolbox for pose estimation based on PyTorch. It is a part of the OpenMMLab project.

The main branch works with PyTorch 1.8+.

https://user-images.githubusercontent.com/15977946/124654387-0fd3c500-ded1-11eb-84f6-24eeddbf4d91.mp4

Major Features

Support diverse tasks

We support a wide spectrum of mainstream pose analysis tasks in current research community, including 2d multi-person human pose estimation, 2d hand pose estimation, 2d face landmark detection, 133 keypoint whole-body human pose estimation, 3d human mesh recovery, fashion landmark detection and animal pose estimation. See Demo for more information.
Higher efficiency and higher accuracy

MMPose implements multiple state-of-the-art (SOTA) deep learning models, including both top-down & bottom-up approaches. We achieve faster training speed and higher accuracy than other popular codebases, such as HRNet. See benchmark.md for more information.
Support for various datasets

The toolbox directly supports multiple popular and representative datasets, COCO, AIC, MPII, MPII-TRB, OCHuman etc. See dataset_zoo for more information.
Well designed, tested and documented

We decompose MMPose into different components and one can easily construct a customized pose estimation framework by combining different modules. We provide detailed documentation and API reference, as well as unittests.

What's New

Release RTMW3D, a real-time model for 3D wholebody pose estimation.
Release RTMO, a state-of-the-art real-time method for multi-person pose estimation.
Release RTMW models in various sizes ranging from RTMW-m to RTMW-x. The input sizes include 256x192 and 384x288. This provides flexibility to select the right model for different speed and accuracy requirements.
Support inference of PoseAnything. Web demo is available here.
Support for new datasets:
- (ICCV 2015) 300VW
Welcome to use the MMPose project. Here, you can discover the latest features and algorithms in MMPose and quickly share your ideas and code implementations with the community. Adding new features to MMPose has become smoother:
- Provides a simple and fast way to add new algorithms, features, and applications to MMPose.
- More flexible code structure and style, fewer restrictions, and a shorter code review process.
- Utilize the powerful capabilities of MMPose in the form of independent projects without being constrained by the code framework.
- Newly added projects include:
  - RTMPose
  - RTMO
  - RTMPose3D
  - PoseAnything
  - YOLOX-Pose
  - MMPose4AIGC
  - Simple Keypoints
  - Just Dance
  - Uniformer
- Start your journey as an MMPose contributor with a simple example project, and let's build a better MMPose together!

January 4, 2024: MMPose v1.3.0 has been officially released, with major updates including:
- Support for new datasets: ExLPose, H3WB
- Release of new RTMPose series models: RTMO, RTMW
- Support for new algorithm PoseAnything
- Enhanced Inferencer with optional progress bar and improved affinity for one-stage methods
Please check the complete release notes for more details on the updates brought by MMPose v1.3.0!

0.x / 1.x Migration

MMPose v1.0.0 is a major update, including many API and config file changes. Currently, a part of the algorithms have been migrated to v1.0.0, and the remaining algorithms will be completed in subsequent versions. We will show the migration progress in this Roadmap.

If your algorithm has not been migrated, you can continue to use the 0.x branch and old documentation.

Installation

Please refer to installation.md for more detailed installation and dataset preparation.

Getting Started

We provided a series of tutorials about the basic usage of MMPose for new users:

For the basic usage of MMPose:
For developers who wish to develop based on MMPose:
For researchers and developers who are willing to contribute to MMPose:
- Contribution Guide
For some common issues, we provide a FAQ list:
- FAQ

Model Zoo

Results and models are available in the README.md of each method's config directory. A summary can be found in the Model Zoo page.

Supported algorithms: