Full Professor, University of Electronic Science and Technology of China
1 paper at NeurIPS 2025
A novel visual token pruning method that jointly maximizes both the saliency and coverage of the selected visual tokens to better preserve semantic completeness.