CRAG-MM

CRAG-MM is a comprehensive benchmark containing 6.5K (image, question, answer) triplets and 2K visual-based multi-turn conversations across 13 domains, used to evaluate Multi-Modal Retrieval-Augmented Generation in wearable device scenarios.

Papers using CRAG-MM (3)

Pixel-Grounded Retrieval for Knowledgeable Large Multimodal Models2026CRAG-MM: Multi-modal Multi-turn Comprehensive RAG Benchmark2025DB3 Team's Solution For Meta KDD Cup' 252025