Datasets
Image + TextScene-text recognition (artistic)2025

ATTW

Artistic Text in the Wild — recognition of stylized scene text

16,627
Artistic instances
7
Source datasets
5
Challenge types
Word-level
Annotation
Samples
Biến thiên hình học mạnh
Nhiễu nền
Đa dạng về font chữ
Hiệu ứng Bolden / 3D
Hiệu ứng Neon / Reflection
About

Artistic Text in the Wild (ATTW) is a benchmark for recognising decorative, stylized and hand-crafted lettering — logos, storefront branding and word-art — where text departs from regular fonts through warping, ornamentation, custom glyphs and heavy background blending. It comprises 16,627 artistic word instances curated from seven established scene-text datasets (WordArt, TotalText, VinText, SignboardText, BKAI-Text, ICDAR-2013, CUTE80) and captures real-world interferences — uneven lighting, clutter, noise, overlapping characters and out-of-vocabulary brand names — grouped along five recurring challenge axes: strong geometric deformation, cluttered background, diverse font styles, bolden/3D effects and neon/reflection effects. ATTW was first introduced at ICDAR 2025 (https://doi.org/10.1007/978-3-032-04630-7_17) and extended in IJDAR 2025 (https://doi.org/10.1007/s10032-025-00565-3). The samples below are a small preview; the full dataset is available on request.