MMDR-Bench

MMDR-Bench (Multi-Modal Dialogue Reasoning Benchmark) is a dataset containing 300 complex multi-turn dialogue scenarios designed to evaluate models on dimensions such as visual entity tracking and reasoning depth.

Papers using MMDR-Bench (3)

CLEAR: Unlocking Generative Potential for Degraded Image Understanding in Unified Multimodal Models2026Can Mllms "read" What Is Missing?2026ContextualLVLM-Agent: A Holistic Framework for Multi-Turn Visually-Grounded Dialogue and Complex Instruction Following2025