Reproduce “Learning Local Shape Descriptors from Part Correspondences With Multi-view Convolutional Networks” 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

Haibin Huang, Evangelos Kalogerakis, Siddhartha Chaudhuri, Duygu Ceylan, Vladimir G. Kim, Ersin Yumer

The authors introduce a learned local descriptor that represents each surface point with rendered neighborhood views at multiple scales, then trains on automatically generated correspondences to bring matching points together. On BHCP’s no-symmetry benchmark, it averages 58.5% correspondence accuracy, versus 40.6% for 3DMatch and 50.5% for shape-context (SC). Scan-to-model examples require no scan-specific fine-tuning, but are qualitative.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.