DOSE: Data Selection for Multi-Modal LLMs via Off-the-Shelf Models

TL;DR

This paper introduces DOSE, a data selection method using off-the-shelf pretrained models to improve multimodal vision-language models by filtering high-quality, diverse data without additional training.

Contribution

Proposes a novel data filtering approach with off-the-shelf models that enhances multimodal model training efficiency and effectiveness without task-specific fine-tuning.

Findings

Models trained on DOSE-filtered data match or outperform those trained on full datasets.

The approach improves data diversity and reduces computational costs.

Experiments validate the method's scalability and effectiveness.

Abstract

High-quality and diverse multimodal data are essential for improving vision-language models (VLMs), yet existing datasets often contain noisy, redundant, and poorly aligned samples. To address these problems, data filtering is commonly used to enhance the efficiency and performance of multimodal learning, but it introduces extra computational cost because filtering models are usually trained on the same data they are meant to screen. To reduce this cost, we study DOSE, which explores whether off-the-shelf pretrained models that have never seen the target data can be used to select training samples for larger and stronger multimodal models without any task-specific training. Even without fine-tuning, these models can effectively assess text quality and image-text alignment to guide data selection. Based on this, we build a joint quality-alignment distribution and apply adaptive weighted sampling to select informative samples while maintaining long-tail diversity. This approach enhances data diversity, enabling models trained on DOSE-filtered data to match or surpass those trained on the full dataset on standard VQA and math benchmarks. Extensive experiments demonstrate its effectiveness, efficiency, and scalability.