WAON: Large-Scale Japanese Image-Text Pair Dataset for Improving Model Performance on Japanese Cultural Tasks

TL;DR

WAON is a large-scale Japanese image-text dataset designed to enhance vision-language models' performance on Japanese cultural tasks, addressing data scarcity and improving benchmark results.

Contribution

The paper introduces WAON, the largest Japanese image-text dataset, and WAON-Bench, a curated benchmark, enabling better fine-tuning and evaluation for Japanese cultural AI applications.

Findings

Fine-tuning on WAON improves Japanese cultural task performance.

WAON achieves state-of-the-art results among comparable models.

The dataset construction pipeline enhances quality through filtering and deduplication.

Abstract

Contrastive pre-training on large-scale image-text pair datasets has driven major advances in vision-language representation learning. Recent work shows that pretraining on global data followed by language or culture specific fine-tuning is effective for improving performance in target domains. With the availability of strong open-weight multilingual models such as SigLIP2, this paradigm has become increasingly practical. However, for Japanese, the scarcity of large-scale, high-quality image-text pair datasets tailored to Japanese language and cultural content remains a key limitation. To address this gap, we introduce WAON, the largest Japanese image-text pair dataset constructed from Japanese web content in Common Crawl, containing approximately 155 million examples. Our dataset construction pipeline employs filtering and deduplication to improve dataset quality. To improve the quality and reliability of evaluation on Japanese cultural tasks, we also construct WAON-Bench, a manually curated benchmark for Japanese cultural image classification comprising 374 classes, which addresses issues in the existing benchmark such as category imbalance and label-image mismatches. Our experiments demonstrate that fine-tuning on WAON improves model performance on Japanese cultural benchmarks more efficiently than existing datasets, achieving state-of-the-art results among publicly available models of comparable architecture. We release our dataset, model, and code.