一作ACL 2026Findings· Findings of ACL
EMC ompress: Video- LLM s with Endomorphic Multimodal Compression
Zheyu Fan · Jiateng Liu · Yuji Zhang · Zihan Wang · Yi R. Fung · Manling Li · Heng Ji
摘要
Video-LLMs face a fundamental tension in long-video reasoning: static, sparse frame sampling either dilutes evidence across task-irrelevant segments at significant cost or misses fine-grained temporal semantics altogether. We propose a novel, cognitively-inspired task — Endomorphic Multimodal Com...