Reproduce “Matisse: Evidence-Space Reasoning for Active 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
Xihang Yu, Kaichen Zhou, Lorenzo Shaikewitz, Clément Jambon, Xiao Zhan, Rajat Talak, Luca Carlone
Matisse uses pretrained 3D generator evidence to guide next-view acquisition and keyframe selection, scoring scene tokens that include generatively completed geometry. On GSO30, it reports 54.167 mm bidirectional Chamfer distance, versus 62.078 mm for the strongest active baseline. In a separate spiral-trajectory experiment, it reaches comparable fitted distance with 7 selected views versus 48 sequentially processed under the reported setups.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.