Paper · arXiv 2408.16357
Law of Vision Representation in MLLMs
Shijia Yang, Bohan Zhai, Quanzeng You, Jianbo Yuan, Hongxia Yang, Chenfeng Xu
Correlation between cross-modal alignment and vision representation improves performance in multimodal large language models, enabling identification and training of optimal vision representation with reduced computational cost.
Abstract
We present the "Law of Vision Representation" in multimodal large language models (MLLMs). It reveals a strong correlation between the combination of cross-modal alignment, correspondence in vision representation, and MLLM performance. We quantify the two factors using the cross-modal Alignment and Correspondence score (AC score). Through extensive experiments involving thirteen different vision representation settings and evaluations across eight benchmarks, we find that the AC score is linearly correlated to model performance. By leveraging this relationship, we are able to identify and train the optimal vision representation only, which does not require finetuning the language model every time, resulting in a 99.7% reduction in computational cost.