The Devil Is in the Details: Delving Into Unbiased Data Processing for Human Pose Estimation

Being a fundamental component in training and inference, data processing has not been systematically considered in human pose estimation community, to the best of our knowledge. In this paper, we focus on this problem and find that the devil of human pose estimation evolution is in the biased data processing. Specifically, by investigating the standard data processing in state-of-the-art approaches mainly including coordinate system transformation and keypoint format transformation (i.e., encoding and decoding), we find that the results obtained by common flipping strategy are unaligned with the original ones in inference. Moreover, there is a statistical error in some keypoint format transformation methods. Two problems couple together, significantly degrade the pose estimation performance and thus lay a trap for the research community. This trap has given bone to many suboptimal remedies, which are always unreported, confusing but influential. By causing failure in reproduction and unfair in comparison, the unreported remedies seriously impedes the technological development. To tackle this dilemma from the source, we propose Unbiased Data Processing (UDP) consist of two technique aspect for the two aforementioned problems respectively (i.e., unbiased coordinate system transformation and unbiased keypoint format transformation). Base on UDP, we wipe out the trap by giving out a deep insight of the existing biased data processing pipeline, whose origin, effects and some confusing remedies are thoroughly studied. Besides, as a model-agnostic approach and a superior solution, UDP successfully pushes the performance boundary of human pose estimation. For example on COCO test-dev set, UDP promotes top-down method HRNet-W32-256×192 by 1.7 AP (73.5 to 75.2) for free and promotes bottom-up methods HRNet-W32-512×512 by 2.7 AP with an acceleration of 6.1 times. The HRNet-W48-384×288 equipped with UDP achieves 76.5 AP and sets a new state-of-the-art for human pose estimation. As a meaningful milestone for pursuing high performance human pose estimation, UDP has been the key base of the winner in 2020 COCO Keypoint Detection Challenge. The code is public available for reference.