Text-Guided Vector Graphics Customization
DescriptionVector graphics are widely used in digital art and highly favored by designers due to their scalability and layer-wise topological properties. However, creating and editing vector graphics require creativity and design expertise, making the process time-consuming. In this paper, we proposes a novel pipeline to generate high-quality customized vector graphics based on textual prompts while preserving the given exemplar SVG's properties and layer-wise information. Our method leverages the power of large pre-trained models for language and visual understanding. By fine-tuning the cross-attention layers of the Stable Diffusion model, we generate customized raster images based on textual prompts. We introduce a semantic-based path alignment method to preserve and transform important paths from the exemplar SVG for SVG initialization. Furthermore, we optimize path parameters using both image-level and vector-level losses to ensure smooth shape deformation while fitting the customized raster image. We evaluate our method using multiple metrics from vector-level, image-level and text-level perspectives. The evaluation results demonstrate the effectiveness of the proposed pipeline in generating diverse vector graphics customizations with high-quality.
|Effective start/end date||1/09/23 → …|