GlyphAtlas: Disentangling Stroke Detail from Layout for Scene Text Synthesis
GlyphAtlas gives small text a detailed stroke reference that does not shrink with its placement in an image.
Existing methods use a pre-rendered glyph image to guide both character structure and layout, limiting stroke detail in small or crowded regions. GlyphAtlas keeps that image as a spatial anchor and adds separate atlas cells containing character or word references. Each text region draws detail only from its corresponding cells. The authors report accurate synthesis of small, dense and complex text, transfer to languages unseen during task-specific training, and leading performance in scenes with many text regions.
HF Daily Papers' note
Existing methods use a pre-rendered glyph image to guide both character structure and layout, limiting stroke detail in small or crowded regions. GlyphAtlas keeps that image as a spatial anchor and adds separate atlas cells containing character or word references. Each text region draws detail only from its corresponding cells. The authors report accurate synthesis of small, dense and complex text, transfer to languages unseen during task-specific training, and leading performance in scenes with many text regions.
HF Daily Papers' note
score 4