The authors built a diffusion transformer that jointly refines compressed representations of object shape and per-image camera pose.
Jointly generating object shapes and camera poses improves sparse-view reconstruction
GenSfM combines 3D diffusion with camera estimation, improving geometry and novel-view rendering from four uncalibrated images on two object benchmarks.
Chinese Tech
Haobin Duan · Binbin Huang · Mochu Xiang · Yiqun Zhao · Zibo Zhao · Yi Ma · +1 more
The University of Hong Kong · Transengram · Tencent
Research Digest··2 min read
Duan et al. introduce GenSfM, a system that estimates a complete 3D object and the cameras that photographed it from a few images, without supplied camera calibration.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.