TY  - RPRT
TI  - Caption First, VQA Second: Knowledge Density, Not Task Format, Drives Multimodal Scaling
AU  - Hongjian Zou
AU  - Yue Ge
AU  - Qi Ding
AU  - Yixuan Liao
AU  - Xiaoxin Chen
PY  - 2026
UR  - https://arxiv.org/abs/2604.13054
ID  - 2604.13054
ER  -