DJW cdece0b32a 第一次提交 9 mesiacov pred
..
coco_wholebody_face cdece0b32a 第一次提交 9 mesiacov pred
lapa cdece0b32a 第一次提交 9 mesiacov pred
wflw cdece0b32a 第一次提交 9 mesiacov pred
README.md cdece0b32a 第一次提交 9 mesiacov pred

README.md

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-WholeBody-Face Dataset

Results on COCO-WholeBody-Face val set

Model Input Size NME Details and Download
RTMPose-m 256x256 0.0466 rtmpose_coco_wholebody_face.md

WFLW Dataset

Results on WFLW dataset

Model Input Size NME Details and Download
RTMPose-m 256x256 4.01 rtmpose_wflw.md

LaPa Dataset

Results on LaPa dataset

Model Input Size NME Details and Download
RTMPose-m 256x256 1.29 rtmpose_lapa.md