StrokeNUWA: Tokenizing Strokes for Vector Graphic Synthesis
This addresses the limitation of raster-based tokenization in visual synthesis for AI and graphics applications, offering a novel approach with significant performance improvements.
The paper tackles the problem of using LLMs for visual synthesis by proposing stroke tokens as a vector graphics representation, which surpasses traditional methods in vector graphic generation metrics and achieves up to a 94x speedup in inference with a 6.9% SVG code compression ratio.
To leverage LLMs for visual synthesis, traditional methods convert raster image information into discrete grid tokens through specialized visual modules, while disrupting the model's ability to capture the true semantic representation of visual scenes. This paper posits that an alternative representation of images, vector graphics, can effectively surmount this limitation by enabling a more natural and semantically coherent segmentation of the image information. Thus, we introduce StrokeNUWA, a pioneering work exploring a better visual representation ''stroke tokens'' on vector graphics, which is inherently visual semantics rich, naturally compatible with LLMs, and highly compressed. Equipped with stroke tokens, StrokeNUWA can significantly surpass traditional LLM-based and optimization-based methods across various metrics in the vector graphic generation task. Besides, StrokeNUWA achieves up to a 94x speedup in inference over the speed of prior methods with an exceptional SVG code compression ratio of 6.9%.