Fetching the paper…

FrameFusion: Combining Similarity and Importance for Video Token Reduction on Large Vision Language Models · Around