Action-conditioned contrastive learning for 3D human pose and shape estimation in videos

Open Webpage

Inpyo Song, Moonwook Ryu, Jangwon Lee

Published: 01 Jan 2024, Last Modified: 08 Nov 2024Comput. Vis. Image Underst. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Leverages human action information for 3D human mesh recovery.•Simple yet effective approach which can adopted to various models.•Improves accuracy without compromising temporal consistency.•Enhances mesh recovery without extra computational costs for inference.