Command Palette
Search for a command to run...
Jiang Tao ; Lu Peng ; Zhang Li ; Ma Ningsheng ; Han Rui ; Lyu Chengqi ; Li Yining ; Chen Kai
Abstract
Recent studies on 2D pose estimation have achieved excellent performance onpublic benchmarks, yet its application in the industrial community stillsuffers from heavy model parameters and high latency. In order to bridge thisgap, we empirically explore key factors in pose estimation including paradigm,model architecture, training strategy, and deployment, and present ahigh-performance real-time multi-person pose estimation framework, RTMPose,based on MMPose. Our RTMPose-m achieves 75.8% AP on COCO with 90+ FPS on anIntel i7-11700 CPU and 430+ FPS on an NVIDIA GTX 1660 Ti GPU, and RTMPose-lachieves 67.0% AP on COCO-WholeBody with 130+ FPS. To further evaluateRTMPose's capability in critical real-time applications, we also report theperformance after deploying on the mobile device. Our RTMPose-s achieves 72.2%AP on COCO with 70+ FPS on a Snapdragon 865 chip, outperforming existingopen-source libraries. Code and models are released athttps://github.com/open-mmlab/mmpose/tree/1.x/projects/rtmpose.
Code Repositories
Benchmarks
| Benchmark | Methodology | Metrics |
|---|---|---|
| 2d-human-pose-estimation-on-coco-wholebody-1 | RTMPose | WB: 65.3 body: 71.4 face: 88.9 foot: 69.2 hand: 59.0 |
| 2d-human-pose-estimation-on-human-art | RTMPose-s | AP: 0.311 |
| 2d-human-pose-estimation-on-human-art | RTMPose-l | AP (gt bbox): 0.753 Validation AP: 83.5 |
| pose-estimation-on-ochuman | RTMPose(RTMPose-l, GT bounding boxes) | Test AP: 80.3 Validation AP: 80.5 |
Build AI with AI
From idea to launch — accelerate your AI development with free AI co-coding, out-of-the-box environment and best price of GPUs.