SketchFlow generates vector sketches from text using Optimal Transport and flow matching

Researchers have introduced SketchFlow, a new generative framework for creating vector sketches from text prompts. This method utilizes Optimal Transport theory and flow matching to map text concepts directly within the CLIP latent space. To handle the gap between text and sketch features, SketchFlow injects noise into category embeddings to form a Gaussian Mixture Model prior, which is then mapped to sketch features using an Optimal Transport Conditional Flow Matching model. A hybrid diffusion decoder, combining 1D U-Net and Transformer architectures, generates the final stroke trajectories. Experiments show SketchFlow surpasses existing methods in visual quality and human-like drawing styles, also demonstrating zero-shot synthesis capabilities for unseen concepts and semantic interpolations. AI

IMPACT This research advances text-to-image generation by enabling zero-shot vector sketch synthesis with human-like styles.

RANK_REASON The cluster contains a research paper detailing a new generative model. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• 1D U-Net
• Flow Matching
• Gaussian mixture model
• Optimal Transport
• QUICKDRAW
• SketchFlow
• Transformer

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →