1
EarlyTom: Early Token Compression Completes Fast Video Understanding
CVPR 2026最新研究：通过早期Token压缩实现快速视频理解，颠覆传统视频处理效率瓶颈。
arXiv:2605.30010v1 Announce Type: new Abstract: Video large language models (Video-LLMs) have demonstrated strong capabilities in video understanding …