4D-DRESS: A 4D Dataset of Real-world Human Clothing with Semantic Annotations

4D-DRESS is the first 4D dataset providing real-world clothed humans with semantic annotations. This dataset contains 4D textured and semantic scans of 64 real-world human outfits in tight and loose garments, with more than 520 motion sequences and 78K scan frames in total. Together with the registered SMPL(-X) body models and captured multi-view images with pixel annotations, 4D-DRESS builds the foundation for future works in human parsing and reconstruction, semantic human avatar learning, realistic cloth reconstruction and simulation.