DreamUV: Unwrap Artist-like UV by End-to-End Flow Matching

DreamUV is an end-to-end learning framework designed for UV parameterization in 3D content creation, addressing the gap between geometric distortion objectives and artistic preferences. It formulates UV unwrapping as a generative Flow Matching problem, employing a boundary-aware training strategy and a Model-in-the-Loop Finetuning scheme to enhance seam geometry and account for discretization errors. Evaluated on a large-scale dataset of artist-authored UVs, DreamUV achieves superior boundary straightness and axis-aligned island tightness compared to classical and learning-based methods, making it a significant advancement for practitioners seeking to produce artist-like UV layouts efficiently.