Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Rethinking on Multi-Stage Networks for Human Pose Estimation

About

Existing pose estimation approaches fall into two categories: single-stage and multi-stage methods. While multi-stage methods are seemingly more suited for the task, their performance in current practice is not as good as single-stage methods. This work studies this issue. We argue that the current multi-stage methods' unsatisfactory performance comes from the insufficiency in various design choices. We propose several improvements, including the single-stage module design, cross stage feature aggregation, and coarse-to-fine supervision. The resulting method establishes the new state-of-the-art on both MS COCO and MPII Human Pose dataset, justifying the effectiveness of a multi-stage architecture. The source code is publicly available for further research.

Wenbo Li, Zhicheng Wang, Binyi Yin, Qixiang Peng, Yuming Du, Tianzi Xiao, Gang Yu, Hongtao Lu, Yichen Wei, Jian Sun• 2019

Related benchmarks

TaskDatasetResultRank
Human Pose EstimationCOCO (test-dev)
AP78.1
408
2D Human Pose EstimationCOCO 2017 (val)
AP75.9
386
Human Pose EstimationMPII (test)
Shoulder PCK97.1
314
Multi-person Pose EstimationCOCO (test-dev)
AP77.1
101
Multi-person Pose EstimationCOCO 2017 (test-dev)
AP78.1
99
Keypoint DetectionMS COCO 2017 (test-dev)
AP78.1
43
Human Keypoint DetectionCOCO
AP78.1
30
Pose EstimationCOCO (test)
AP76.1
28
Human Keypoint DetectionMS COCO (test-dev)
AP78.1
19
Pose EstimationCOCO 2017 (test-challenge)
AP76.4
6
Showing 10 of 11 rows

Other info

Code

Follow for update