Garments2Look

Garments2Look is a CVPR 2026 dataset comprising 80,000 many-garments-to-one-look triplets across 40 major and 300+ fine-grained categories, built jointly by The Hong Kong Polytechnic University and Huhu AI Inc. It is the first large-scale benchmark for outfit-level virtual try-on with multi-reference garment conditioning, including fabric, layering order, and dressing-technique annotations.

First mentioned: May 2026