RTMPose
Recent studies on 2D pose estimation have achieved excellent performance on public benchmarks, yet its application in the industrial community still suffers from heavy model parameters and high latency.
In order to bridge this gap, we empirically study five aspects that affect the performance of multi-person pose estimation algorithms: paradigm, backbone network, localization algorithm, training strategy, and deployment inference, and present a high-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 an Intel i7-11700 CPU and 430+ FPS on an NVIDIA GTX 1660 Ti GPU, and RTMPose-l achieves 67.0% AP on COCO-WholeBody with 130+ FPS, outperforming existing open-source libraries.
To further evaluate RTMPose's capability in critical real-time applications, we also report the performance after deploying on the mobile device.
Results and Models
COCO Dataset
Results on COCO val2017 with detector having human AP of 56.4 on COCO val2017 dataset
MPII Dataset
| Model |
Input Size |
PCKh@0.5 |
PCKh@0.1 |
Details and Download |
| RTMPose-m |
256x256 |
0.907 |
0.348 |
rtmpose_mpii.md |
CrowdPose Dataset
Results on CrowdPose test with YOLOv3 human detector
Human-Art Dataset
Results on Human-Art validation dataset with detector having human AP of 56.2 on Human-Art validation dataset
Results on Human-Art validation dataset with ground-truth bounding-box
eeb5095331