TY  - RPRT
TI  - Top-Down Compression: Revisit Efficient Vision Token Projection for Visual Instruction Tuning
AU  - Bonan li
AU  - Zicheng Zhang
AU  - Songhua Liu
AU  - Weihao Yu
AU  - Xinchao Wang
PY  - 2025
UR  - https://arxiv.org/abs/2505.11945
ID  - 2505.11945
ER  -