Shanghai Jiao Tong University, Machine Vision and Intelligence Group(MVIG)


Alpha Pose is an accurate multi-person pose estimation system. It is the first open-sourced system that can achieve 70+ mAP (72.3 mAP) on COCO dataset and 80+ mAP (82.1 mAP) on MPII dataset. To associate poses that indicates the same person across frames, we also provide an efficient online pose tracker called Pose Flow. It is also the first open-sourced online pose tracker that can both satisfy 60+ mAP (66.5 mAP) and 50+ MOTA (58.3 MOTA) on PoseTrack Challenge dataset.

Developed and maintained by Hao-Shu Fang, Jiefeng Li, Yuliang Xiu, Ruiheng Chang and Cewu Lu(corresponding authors)


  • Accurate multi-person keypoint detection.
  • Input: Image, video, image list.
  • Output: Basic image + keypoint display/saving (PNG, JPG, AVI, ...), keypoint saving (JSON), supports multiple formats.
  • Available: command-line demo, python and Lua programs
  • OS: Ubuntu


Code and Papers

Our source code is available on Github, and our paper can be downloaded from here.


Please cite these papers if you use AlphaPose:

                   title={RMPE: Regional Multi-person Pose Estimation},
                   author={Fang, Hao-Shu and Xie, Shuqin and Tai, Yu-Wing and Lu, Cewu},

Related Research using AlphaPose:

Adverb Recognition

ADHA: A Benchmark for Recognizing Adverbs describing Human Actions in Videos(arXiv'17)
Bo Pang, Kaiwen Zha, Cewu Lu(corresponding author)

We propose a new task of Action Adverb recongnition. Aphapose helps to improve this task.


Pose Tracker

Pose Flow: Efficient Online Pose Tracking(arXiv'18)
Yuliang Xiu, Jiefeng Li, Haoyu Wang, Yinghong Fang, Cewu Lu(corresponding author)

Our pose tracker achieves state-of-the-art preformance in pose tracking task, and the extra computaion on Alphapose is very minor, requiring 0.01 second per frame only.