Abstract
We integrate smoothing B-splines into a standard differentiable vector graphics (DiffVG) pipeline through linear mapping, and show how this can be used to generate smooth and arbitrarily long paths within image-based deep learning systems. We take advantage of derivative-based smoothing costs for parametric control of fidelity vs. simplicity tradeoffs, while also enabling stylization control in geometric and image spaces. The proposed pipeline is compatible with recent vector graphics generation and vectorization methods. We demonstrate the versatility of our approach with four applications aimed at the generation of stylized vector graphics: stylized space-filling path generation, stroke-based image abstraction, closed-area image abstraction, and stylized text generation.
Our method allows to optimize long and smooth B-Spline curves using DiffVG rasterization pipelines. Applications range from text stylization to image abstraction and vectorization.
Flow chart of our pipeline, all operations are differentiable.
Examples of stylized area filling for a letter “S”. The images on the lower left are used to guide stylization.
Stroke abstraction of Thelonious Monk. Left, using a single stroke and Voronoi with TSP initialzation. Middle, using multiple strokes with multiple key-points along vertical lines. Right, using facial features extracted with MediaPipe.
Rendering stroke abstractions with a spray-like brush. The quintic B-splines together with smoothing on jerk produce smooth motions that tend to slow down where curvature is higher.
Our method produces smooth kinematics that facilitate reproduc- tion with a robot, and the varying width can be used to control brush pressure. Left and center, the robot reproducing portraits. Right, the robot reproducing a stylized area fill.
Area-based image abstraction with quantized colors and image-driven style guidance. The palettes are extracted from the style image. The last row shows increasing stylization weight from left to right.
Monospace font generation. Combining glyphs optimized to fit a triangle, a square and a circle.
Calligram generation. Deforming glyphs to fit animal silhouettes.
BibTeX
@article{NeurSplines-25,
title = {Neural Image Abstraction using Long Smoothing B-Splines},
author = {Daniel Berio and Michael Stroh and Sylvain Calinon and Frederic Fol Leymarie and Oliver Deussen and Ariel Shamir },
journal = {ACM Transactions on Graphics (SIGGRAPH Asia 2025 Conference Proceedings)},
year = {2025},
volume = {44},
Number = {6},
pages = {Accepted},
}