#3d-reconstruction

Sparse 3D capture fails where the camera never looked. NVIDIA Research says ArtiFixer generates hundreds of frames in one pass and beats prior methods by 1-3 dB PSNR on common benchmarks.

A Hacker News discussion highlighted LoGeR, a Google DeepMind and UC Berkeley project that uses hybrid memory to scale dense 3D reconstruction across extremely long videos without post-hoc optimization.