X Scaling Up Expressive Human Pose and Shape Estimation Supplementary Material

Due to space constraints in the main paper, we elaborate the following here: additional details of the 32 datasets, including useful links to find their license statements and other ethics concerns in Sec. B.1 Dataset Descriptions This section describes the 32 datasets we study. Note that all these are public academic datasets, each holding a license. We follow the common practice to use them in our non-commercial research and refer readers to their homepages or papers for more details regarding licenses and their policies to ensure personal information protection. It features accurate SMPL annotations and 60 video sequences captured in diverse environments. We follow the official definition of train, val, and test splits. AGORA [34] (Figure 1b) is a synthetic dataset, rendered with high-quality human scans and realistic 3D scenes. It consists of 4240 textured human scans with diverse poses and appearances, each fitted with accurate SMPL-X annotations.