Reproduce “SPOON: Towards Coherent Compositional 3D Scene Generation from Uncalibrated Multi-view Images” 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
Guibiao Liao, Mochu Xiang, Heng Li, Ken Deng, Zijie Wang, Guanbin Li, Ping Tan, Shenghua Gao and 1 more
SPOON coordinates poses across uncalibrated views to generate coherent 3D scenes. It uses reconstructed geometry to guide pose generation, select the most observable view per object, and refine object placements and camera poses together. On ARSG-110K with four views, versus MV-SAM3D, scene-level Chamfer Distance falls from 0.063 to 0.055 and bounding-box IoU rises from 0.585 to 0.662; this is a matched-setting result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.