Reproduce “Dense4D: Pixel-Level 4D Human Correspondence” 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

Rıza Alp Güler

Dense4D tracks surface points across motion by mapping pixels from multiple frames into a shared reference-time pose, matching coordinates, and using predicted depth for 3D tracks. On Panoptic Studio Humans, it scores 38.4 AJ versus 17.0 for TAPIP3D; on 4D-DRESS, its mean one-sided point-to-scan distance is 0.86 cm. Scan proximity does not prove point identity; Panoptic tracking labels are reconstruction-derived.

Summary by alphaXiv.

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