
Video-LMMs struggle to correctly comprehend complex videos, indicating their weak reasoning and lack of robustness to textual user queries. A training-free Dual-Step Contextual Prompting (DSCP) technique can effectively enhance the performance of existing Video-LMMs.


coremsg

comprehensive-evaluation-of-video-large-multi-modal-models-reasoning-and-robustness-capabilities-in-complex-real-world-scenarios


Comprehensive Evaluation of Video Large Multi-modal Models' Reasoning and Robustness Capabilities in Complex Real-World Scenarios
