CVPR2025

From Head to Tail: Towards Balanced Representation in Large Vision-Language Models through Adaptive Data Calibration

Mingyang Song, Xiaoye Qu, Jiawei Zhou, Yu Cheng

Abstract

Experiments Ablation Study Ablation study on data rebalancing combinations. Ablation study on data synthesis methods. Visualization Qualitative comparison between LLaVA 1.5 and LLaVA 1.5 w/ ADR Comparison between the original instruction-tuning (IT) data and our synthesized IT data.