Date: July 2017.
Source: ACM Transactions on Graphics, Vol. 36, Issue 4, Article 73.
Research Summary: ClothCap addresses the problem of capturing regular clothing on fully dressed people in motion in 3D/4D, including cases where a person wears multiple garments at once. The method introduces a multi-part 3D model of clothed bodies, built on the SMPL body model, that automatically segments each garment, estimates the body’s shape and pose underneath the clothing, and tracks how each garment deforms over time. To start, garments and their motion are estimated from real-world 4D scans, high-resolution 3D captures of a subject in motion at 60 frames per second. The results demonstrate a methodology to capture real-world 3D/4D data of a clothed person in motion, extract their clothing, and retarget it to new 3D body shapes, a foundational step toward virtual try-on.
3dMD’s Role: 3dMD’s 4D real-world body data is the dense-surface 3D-shape foundation of ClothCap. ClothCap’s own methods identify the same active stereo capture system used in Dyna, the 3dMD dataset behind Dynamic FAUST, and MPI-IS carried that same system forward into CAPE (2020).
Article: ClothCap: Seamless 4D Clothing Capture and Retargeting.
Authors: Gerard Pons-Moll, Sergi Pujades, Sonny Hu, Michael J Black. Max Planck Institute for Intelligent Systems, Perceiving Systems