Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wu, Biao, Zhong, Yiwu, Fang, Meng, Chen, Ling
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2604.16979
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917418009362432
author Wu, Biao
Zhong, Yiwu
Fang, Meng
Chen, Ling
author_facet Wu, Biao
Zhong, Yiwu
Fang, Meng
Chen, Ling
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2604_16979
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DOSE: Data Selection for Multi-Modal LLMs via Off-the-Shelf Models
Wu, Biao
Zhong, Yiwu
Fang, Meng
Chen, Ling
Computer Vision and Pattern Recognition
Computation and Language
F.2.2; I.2.7
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.
title DOSE: Data Selection for Multi-Modal LLMs via Off-the-Shelf Models
topic Computer Vision and Pattern Recognition
Computation and Language
F.2.2; I.2.7
url https://arxiv.org/abs/2604.16979