VisualWebInstruct: Scaling up Multimodal Instruction Data through Web Search

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jia, Yiming, Li, Jiachen, Yue, Xiang, Li, Bo, Nie, Ping, Zou, Kai, Chen, Wenhu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910876685041664
author Jia, Yiming
Li, Jiachen
Yue, Xiang
Li, Bo
Nie, Ping
Zou, Kai
Chen, Wenhu
author_facet Jia, Yiming
Li, Jiachen
Yue, Xiang
Li, Bo
Nie, Ping
Zou, Kai
Chen, Wenhu
contents Vision-Language Models have made significant progress on many perception-focused tasks. However, their progress on reasoning-focused tasks remains limited due to the lack of high-quality and diverse training data. In this work, we aim to address the scarcity of reasoning-focused multimodal datasets. We propose VisualWebInstruct, a novel approach that leverages search engines to create a diverse and high-quality dataset spanning multiple disciplines, including mathematics, physics, finance, and chemistry, etc. Starting with a meticulously selected set of 30,000 seed images, we employ Google Image Search to identify websites containing similar images. We collect and process HTML data from over 700K unique URLs. Through a pipeline of content extraction, filtering, and synthesis, we construct a dataset of approximately 900K question-answer (QA) pairs, with 40% consisting of visual QA pairs and the remaining comprising text-based QA pairs. Models fine-tuned on VisualWebInstruct demonstrate significant performance improvements: (1) fine-tuning on Llava-OV results in 10-20 absolute points improvement across benchmarks, and (2) fine-tuning from MAmmoTH-VL yields a 5 absolute points gain across benchmarks. Our best model, MAmmoTH-VL2, achieves state-of-the-art performance within the 10B parameter class on MMMU-Pro (40.7), MathVerse (42.6), and DynaMath (55.7). These results highlight the effectiveness of our dataset in enhancing the reasoning capabilities of vision-language models for complex multimodal tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2503_10582
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle VisualWebInstruct: Scaling up Multimodal Instruction Data through Web Search
Jia, Yiming
Li, Jiachen
Yue, Xiang
Li, Bo
Nie, Ping
Zou, Kai
Chen, Wenhu
Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Vision-Language Models have made significant progress on many perception-focused tasks. However, their progress on reasoning-focused tasks remains limited due to the lack of high-quality and diverse training data. In this work, we aim to address the scarcity of reasoning-focused multimodal datasets. We propose VisualWebInstruct, a novel approach that leverages search engines to create a diverse and high-quality dataset spanning multiple disciplines, including mathematics, physics, finance, and chemistry, etc. Starting with a meticulously selected set of 30,000 seed images, we employ Google Image Search to identify websites containing similar images. We collect and process HTML data from over 700K unique URLs. Through a pipeline of content extraction, filtering, and synthesis, we construct a dataset of approximately 900K question-answer (QA) pairs, with 40% consisting of visual QA pairs and the remaining comprising text-based QA pairs. Models fine-tuned on VisualWebInstruct demonstrate significant performance improvements: (1) fine-tuning on Llava-OV results in 10-20 absolute points improvement across benchmarks, and (2) fine-tuning from MAmmoTH-VL yields a 5 absolute points gain across benchmarks. Our best model, MAmmoTH-VL2, achieves state-of-the-art performance within the 10B parameter class on MMMU-Pro (40.7), MathVerse (42.6), and DynaMath (55.7). These results highlight the effectiveness of our dataset in enhancing the reasoning capabilities of vision-language models for complex multimodal tasks.
title VisualWebInstruct: Scaling up Multimodal Instruction Data through Web Search
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2503.10582