Meituan Technology Team
May 16, 2024 · Artificial Intelligence
CMIngre: A Cross‑Modal Ingredient‑Level Dataset for Chinese Food Understanding
The CMIngre dataset, created by Meituan’s R&D platform and Tianjin University, offers 8,001 image‑text pairs of 429 Chinese dishes with 95,290 ingredient bounding boxes, enabling fine‑grained ingredient detection and cross‑modal retrieval tasks, and baseline experiments show DINO and CLIP models achieve the strongest performance.
computer visioncross-modal retrievalfood understanding
0 likes · 44 min read