<p>Recognizing artistic text presents significant challenges due to its design by artists and professional designers, characterized by diverse fonts, text effects, layouts, personal styles, and complex backgrounds. Despite its importance, the problem of artistic text recognition remains insufficiently addressed within the research community. To address this gap, we introduce a benchmark dataset, Artistic Text-In-The-Wild (ATTW), consisting of 16,627 diverse instances of artistic text. ATTW serves as a comprehensive and diverse benchmark that captures a wide range of real-world artistic text variations, promoting the development and evaluation of more effective scene text recognition approaches. Furthermore, we propose a novel approach to artistic text recognition that emphasizes the use of skeletal information, positing that such information is essential for enhancing recognition accuracy. Experimental results demonstrate that the skeleton-guided method outperforms existing SOTA methods on the ATTW benchmark, highlighting the robustness and efficacy of the approach in tackling the unique challenges posed by artistic text.</p>