1 paper · 1 filter
Yudong Han, Qingpei Guo, Liyuan Pan +3
The challenge in LLM-based video understanding lies in preserving visual and semantic information in long videos while maintaining a memory-affordable token count. However, redunda…