Abstract
We integrate smoothing B-splines into a standard differentiable vectorgraphics (DiffVG) pipeline through linear mapping, and show how this can beused to generate smooth and arbitrarily long paths within image-based deeplearning systems. We take advantage of derivative-based smoothing costs forparametric control of fidelity vs. simplicity tradeoffs, while also enablingstylization control in geometric and image spaces. The proposed pipeline iscompatible with recent vector graphics generation and vectorization methods. Wedemonstrate the versatility of our approach with four applications aimed at thegeneration of stylized vector graphics: stylized space-filling path generation,stroke-based image abstraction, closed-area image abstraction, and stylizedtext generation.