A Large-scale Dataset for Robust Complex Anime Scene Text Detection

2025-10-09

AI Summary

Existing text detection datasets are primarily designed for natural scenes or document images, rendering them inadequate for anime—where text exhibits high stylistic diversity, irregular layouts, frequent visual ambiguity with symbols/decorations, and pervasive use of handwritten and artistic fonts. To address this gap, we introduce AnimeText, the first large-scale, anime-specific text detection dataset, comprising 735K images and 4.2M annotated text instances. We propose a hierarchical annotation scheme and a hard-negative sampling strategy to comprehensively cover non-linear arrangements, multi-font variations, and severe visual interference, while ensuring compatibility with mainstream deep learning frameworks. Cross-dataset benchmarking demonstrates that models trained on AnimeText significantly outperform those trained on general-purpose datasets on anime text detection tasks, validating AnimeText’s effectiveness and robustness.

Technology Category

• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Computer Vision: Object Detection & Categorization
• Machine Learning: Multimodal Learning

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Current text detection datasets primarily target natural or document scenes, where text typically appear in regular font and shapes, monotonous colors, and orderly layouts. The text usually arranged along straight or curved lines. However, these characteristics differ significantly from anime scenes, where text is often diverse in style, irregularly arranged, and easily confused with complex visual elements such as symbols and decorative patterns. Text in anime scene also includes a large number of handwritten and stylized fonts. Motivated by this gap, we introduce AnimeText, a large-scale dataset containing 735K images and 4.2M annotated text blocks. It features hierarchical annotations and hard negative samples tailored for anime scenarios. %Cross-dataset evaluations using state-of-the-art methods demonstrate that models trained on AnimeText achieve superior performance in anime text detection tasks compared to existing datasets. To evaluate the robustness of AnimeText in complex anime scenes, we conducted cross-dataset benchmarking using state-of-the-art text detection methods. Experimental results demonstrate that models trained on AnimeText outperform those trained on existing datasets in anime scene text detection tasks. AnimeText on HuggingFace: https://huggingface.co/datasets/deepghs/AnimeText

Problem

• Detecting diverse stylized text in anime scenes
• Addressing irregular text arrangements in complex anime visuals
• Overcoming confusion between text and decorative anime elements

Innovation

• Large-scale anime text dataset with 735K images
• Hierarchical annotations for complex anime scenarios
• Hard negative samples for robust text detection