1 paper · 1 filter
Yangfu Li, Hongjian Zhan, Tianyi Chen +2
Existing visual token pruning methods target prompt alignment and visual preservation with static strategies, overlooking the varying relative importance of these objectives across…