Reproduce “ControlFlow3D: Distilling Multi-View Knowledge into Latent Flow Matching for Point Cloud Upsampling” 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

Yuang Liu, Zhi Zuo, Zhengkai Zhao, Lirui Zhang, Pan Gao, Hao Feng, Zhengzhe Liu

ControlFlow3D upsamples sparse point clouds using multi-view images as training-time guidance, then runs at inference from sparse points alone. It learns feature-space flow toward dense geometry, with shared prototypes transferring visual cues without requiring point-to-image correspondences. On ShapeNetPU 4×, its point-to-surface distance is 0.818 versus 1.672 for PUFM; this result is limited to the tested benchmark and protocol.

Summary by alphaXiv.

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