Fetching the paper…

LLaVA-Scissor: Token Compression with Semantic Connected Components for Video LLMs · Around