VisNec: Measuring and Leveraging Visual Necessity for Multimodal Instruction Tuning
This addresses inefficiencies in multimodal AI training for researchers and practitioners by enabling more robust and efficient instruction tuning with reduced data requirements.
The paper tackled the problem of visually redundant and misaligned samples in multimodal instruction tuning by proposing VisNec, a framework that measures visual necessity for data selection, resulting in training on only 15% of data achieving 100.2% of full-data performance and on a smaller dataset surpassing full-data training by 15.8%.
The effectiveness of multimodal instruction tuning depends not only on dataset scale, but critically on whether training samples genuinely require visual reasoning. However, existing instruction datasets often contain a substantial portion of visually redundant samples (solvable from text alone), as well as multimodally misaligned supervision that can degrade learning. To address this, we propose VisNec (Visual Necessity Score), a principled data selection framework that measures the marginal contribution of visual input during instruction tuning. By comparing predictive loss with and without visual context, VisNec identifies whether a training instance is vision-critical, redundant, or misaligned. To preserve task diversity, we combine VisNec with semantic clustering and select high-necessity samples within each cluster. Across 10 downstream benchmarks, training on only 15% of the LLaVA-665K dataset selected by VisNec achieves 100.2% of full-data performance. On the smaller Vision-Flan-186K dataset, our selection not only further reduces data size but also surpasses full-data training by 15.8%. These results demonstrate that measuring and leveraging visual necessity provides an effective solution for both efficient and robust multimodal instruction tuning. Codes and selected subsets will be released upon acceptance.