Reproduce “Multi-View 3D Reconstruction Using V-JEPA Features” 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

Anwar Shamim, Jeffrin Sam

The method reconstructs 3D scenes from a few uncalibrated RGB views, using pretrained V-JEPA features with explicit camera estimation and cross-view geometric reasoning. It requires only RGB at inference, while geometry labels supervise training. On 168 CO3D five-view windows, spatial cameras reduce mean camera ATE versus global regression, but identity poses do better and depth error worsens.

Summary by alphaXiv.

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