Reproduce “ReDT: Resolution-Decoupled Visual Geometry Transformer” 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
Abteen Arab, Guile Wu, Dongfeng Bai
ReDT addresses the cost of high-resolution multi-view 3D reconstruction by letting the network reason over compact learned features before restoring them for dense predictions. Its entropy-guided tokenizer adapts patch sizes, while DIVA projects spatial features into a fixed-size basis for correspondence reasoning. On ETH3D at 2K, ReDT-Ω reports lower CD, runtime, and peak memory than VGGT-Ω; these are benchmark results, not universal guarantees.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.