A dual-channel network based on occlusion feature compensation for human pose estimation

Jiahong Jiang, Nan Xia

Abstract: Highlights•We propose a dual-channel network for occluded human pose estimation.•We propose the OHE-attention to identify occluded keypoints and assign attention weights to the occlusion area.•We propose a feature compensation method to eliminates obstacle features and integrates shared and private human attributes.