JSL3d: Joint subspace learning with implicit structure supervision for 3D pose estimation
作者：
Highlights：
• JSL3d imposes implicit structure representations obtained by the learning-based approach into the inference frame of SR model to capture complex structure relations of human poses.
• The optimization procedure of JSL3d is built on projection and latent spaces, in which the implicit spatial relation consistency of body joints between 2D and 3D spaces is enforced.
• Extensive experiments conducted on well-recognized benchmarks demonstrate the effectiveness of JSL3d.
摘要
•JSL3d imposes implicit structure representations obtained by the learning-based approach into the inference frame of SR model to capture complex structure relations of human poses.•The optimization procedure of JSL3d is built on projection and latent spaces, in which the implicit spatial relation consistency of body joints between 2D and 3D spaces is enforced.•Extensive experiments conducted on well-recognized benchmarks demonstrate the effectiveness of JSL3d.
论文关键词：3D pose estimation,Sparse representation model,Implicit structure supervision,Joint subspace learning
论文评审过程：Received 19 November 2019, Revised 2 July 2022, Accepted 9 August 2022, Available online 10 August 2022, Version of Record 19 August 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2022.108965
