Reproduce “Iris-Z: Scaling One Generative Model Past Specialist Geometry Models” 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

Alejandro Cañada, Chema Garabito

Iris-Z tests whether one generative model can estimate relative depth while retaining image generation. It encodes images and depth with shared components in one diffusion transformer, without a separate depth network. On Marigold V2’s five-set benchmark, Iris-Z-big scores 2.92 versus MoGe-3’s 3.12, leading on three sets; MoGe-3 leads on two, so the result is not universal.

Summary by alphaXiv.

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