Reproduce “HIST3R: Decomposing and Rectifying State Updates with History for Streaming 3D Reconstruction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kenneth Yang, Yu-Chiang Frank Wang
HIST3R improves streaming 3D reconstruction by adjusting both the direction and strength of updates to CUT3R’s recurrent state. It uses recent states to steer update direction and visual differences from history to modulate update strength, without retraining the base model. On ScanNet at 1,000 input frames, it reduces ATE by 23.87% versus TTSA3R, demonstrating a long-sequence pose-estimation gain, not a win across every task or input regime.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.