Justin Valentine, Riley Zilka, Hasti Karamdel, University of Alberta
This is our implementation of "Sequential Sketch Generation via Guided Flow Matching", built upon the Diffusion Policy framework. Our method leverages Flow Matching and classifier-free diffusion guidance to generate human-like sketches while preserving diverse drawing styles and behaviors. We train models on the Quick, Draw! dataset, the TU-Berlin sketch dataset, and the The Sketchy Database.
