Umer Saleem

Position: Video LLMs Must Not Ignore the Pixel Dynamics in Plain Sight

The essence of video lies in pixel dynamics: motion, state transitions, and the flow of visual information across frames. Video Large Language Models (LLMs) have rapidly become the dominant paradigm for video understanding in computer vision, sophisticated multimodal reasoning over complex, long-for…