Bezier Splatting for Fast and Differentiable Vector Graphics Rendering

TL;DR

Bézier Splatting introduces a differentiable vector graphics representation that rasterizes Bézier-curve primitives via 2D Gaussian splatting, enabling an order-of-magnitude speedup over prior methods like DiffVG while delivering higher visual fidelity. It integrates an adaptive pruning and densification mechanism to reallocate curve density in high-error regions, creating a global receptive field that improves optimization. The approach supports open and closed curves, exhibits strong performance across natural and non-photorealistic imagery, and is fully compatible with SVG, facilitating editing and interoperability. Overall, Bézier Splatting offers a scalable, differentiable VG pipeline with practical applicability to high-resolution vectorization and editing tasks.

Abstract

Differentiable vector graphics (VGs) are widely used in image vectorization and vector synthesis, while existing representations are costly to optimize and struggle to achieve high-quality rendering results for high-resolution images. This work introduces a new differentiable VG representation, dubbed Bézier Splatting, that enables fast yet high-fidelity VG rasterization. Bézier Splatting samples 2D Gaussians along Bézier curves, which naturally provide positional gradients at object boundaries. Thanks to the efficient splatting-based differentiable rasterizer, Bézier Splatting achieves 30x and 150x faster per forward and backward rasterization step for open curves compared to DiffVG. Additionally, we introduce an adaptive pruning and densification strategy that dynamically adjusts the spatial distribution of curves to escape local minima, further improving VG quality. Furthermore, our new VG representation supports conversion to standard XML-based SVG format, enhancing interoperability with existing VG tools and pipelines. Experimental results show that Bézier Splatting significantly outperforms existing methods with better visual fidelity and significant optimization speedup.

Figures (16)

• Figure 1: This work proposes Bézier Splatting, a new differentiable vector graphics (VGs) renderer that achieves an order-of-magnitude computational speedup in comparison with the state-of-the-art method DiffVG Li:2020:DVG (tested on a NVIDIA RTX 4090 GPU).
• Figure 2: An illustration of the algorithm flow of Bézier Splatting. It begins by randomly initializing Bézier curves and uniformly sampling Gaussians points along them. These Gaussians are then rasterized into the image, enabling gradient-based computation to optimize parameters of both Bézier curves and Gaussians. Curves with negligible opacity or extremely small shapes are removed, while new curves are adaptively added into areas with high reconstruction error, ensuring curves are placed in areas requiring finer details. $\rightarrow$ forward, $\leftarrow$ backpropagation, $\leftarrow$ error map.
• Figure 3: A qualitative comparison of our method and the state-of-the-art differentiable VG rasterization methods, including DiffVG Li:2020:DVG, LIVE xu2022live, and LIVSS livss.
• Figure 4: Our Bézier Splatting achieves high-quality image vectorization results for various types of images including artworks, cartoons, and natural images. Curve type and count are indicated at the bottom right of each sample.
• Figure 5: As a differentiable VGs renderer, Bézier Splatting can integrate with topology-aware strategies such as layer-wise vectorization xu2022live to further improve compositionality and detail.