Advancing WordArt-Oriented Scene Text Recognition: Datasets and Methods

2026-06-23

AI Summary

This work addresses the challenge of recognizing WordArt—stylized text with highly customized fonts, textures, and layouts—which remains difficult for conventional scene text recognition methods. To this end, the authors propose WATERec, an end-to-end recognition framework that leverages a large-scale synthetic dataset, WATER-S, and a flexible architecture capable of handling arbitrary-shaped inputs and complex typographic layouts. The dataset is constructed through controllable rendering, prompt mining using Qwen3-VL, and Z-Image synthesis techniques. The model employs an adaptable visual encoder coupled with an autoregressive decoder, thereby overcoming limitations imposed by fixed templates. Evaluated on WordArt-Bench, WATERec achieves a state-of-the-art accuracy of 90.40%, substantially outperforming both general-purpose and OCR-specific vision-language models and establishing a new benchmark for irregular artistic text recognition.

Abstract

WordArt (artistic text) features highly customized fonts, textures, and layouts, making WordArt-oriented scene TExt Recognition (WATER) substantially more challenging than general Scene Text Recognition (STR). Existing STR datasets and methods, typically built around regular scene text and fixed-template inputs, struggle to scale to WATER. Thus, we aim to advance this task from both data and model perspectives. On the data side, we construct a 2M synthetic dataset, WATER-S, with the scale improved by hundreds of times compared to existing artistic text data. WATER-S consists of two complementary subsets. One rendered by an upgraded rendering pipeline (SynthWordArt), which provides highly accurate and controllable synthetic WordArt data. The other is generated by combining Qwen3-VL for prompt mining and Z-Image for image synthesis, which improves the coverage of realistic and diverse data. On the model side, we propose WATERec. It adopts an visual encoder supporting arbitrary-shaped inputs and an autoregressive decoder to model complex layouts, structurally breaking the bottleneck of fixed-template STR on WordArt. Experiments show that this architecture outperforms prior STR methods, achieving state-of-the-art performance on irregular texts such as WordArt. Together with WATER-R, carefully reorganized from existing real STR data, our strong baseline with the new synthetic data and model design reaches 90.40% accuracy on WordArt-Bench, surpassing both general-purpose and OCR-specialized vision-language models by a large margin. Code and data are available at https://github.com/YesianRohn/WATER.

Problem

• WordArt
• Scene Text Recognition
• Artistic Text
• Irregular Text
• Text Recognition

Innovation

• WordArt
• scene text recognition
• synthetic dataset
• arbitrary-shaped input
• autoregressive decoder