D-CoDe: Scaling Image-Pretrained VLMs to Video via Dynamic Compression and Question Decomposition.

Yiyang Huang, Yizhou Wang, Yun Fu

Browse the full EMNLP paper archive.