VideoMME

The 'VideoMME' dataset is a benchmark used to evaluate long video understanding in Multi-modal Large Language Models (MLLMs) by assessing their performance on tasks involving multi-modal information from extended video content.

Papers using VideoMME (3)

AdaptToken: Entropy-based Adaptive Token Selection for MLLM Long Video Understanding2026Attend Before Attention: Efficient and Scalable Video Understanding via Autoregressive Gazing2026MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe2025