We are working on two main topics related to 3D:
- reconstructing 3D models from historical photographs and depictions
- mining 3D data for repetitive elements.
We have made several important contribution to 3D data representations with deep learning, that we think are key steps toward these goals. Their main idea is to focus on describing surface deformations, rather than directly considering 3D points or voxels.
Publications:
Learning elementary structures for 3D shape generation and matching, Thé Deprelle, Thibault Groueix, Matthew Fisher, Vladimir G. Kim, Bryan C. Russell and Mathieu Aubry, NeurIPS 2019, PDF, Project page
3D-CODED : 3D Correspondences by Deep Deformation, Thibault Groueix, Matthew Fisher, Vladimir G. Kim, Bryan C. Russell and Mathieu Aubry, ECCV 2018, PDF, Project page
AtlasNet: A Papier-Mâché Approach to Learning Surface Generation, Thibault Groueix, Matthew Fisher, Vladimir G. Kim, Bryan C. Russell and Mathieu Aubry, CVPR 2018, PDF, Project page