RTMPose: Real-Time Models for Multi-Person Pose Estimation

Tao Jiang, Peng Lu, Li Zhang, Ningsheng Ma, Han Lyu, Chengqi Lyu, Yining Li, Kai Chen

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: human pose estimation, real-time, coco, coco-wholebody, crowdpose

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: Empirically explore key factors in pose estimation including paradigm, model architecture, training strategy, and deployment, and present a high-performance real-time multi-person pose estimation pipeline, RTMPose.

Abstract: 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. To bridge this gap, we empirically explore key factors in pose estimation including paradigm, model architecture, training strategy, and deployment, and present a high-performance real-time multi-person pose estimation pipeline, RTMPose. 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-x achieves 65.3% AP on COCO-WholeBody. To further evaluate RTMPose's capability in critical real-time applications, we also report the performance after deploying on the mobile device. Our RTMPose-s model achieves 72.2% AP on COCO with 70+ FPS on a Snapdragon 865 chip, outperforming existing methods used by industrial companies.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 2419