Neural Image Abstraction Using Long Smoothing B-Splines

2025-11-07

AI Summary

This work addresses the challenge of modeling long, smooth paths in deep learning–driven vector graphic generation. We propose a novel method embedding differentiable B-splines into neural rendering pipelines. Our approach features: (1) a derivative-based smoothness loss enabling end-to-end optimization of arbitrarily long paths; (2) a linear mapping that seamlessly integrates B-splines into the DiffVG framework, supporting jointly differentiable geometric and pixel-space rendering; and (3) a fidelity–simplicity trade-off parameter enabling multi-task generation—including stylized fill, stroke abstraction, closed-region simplification, and stylized text rendering. Experiments demonstrate substantial improvements in path smoothness and structural simplicity while preserving high visual fidelity. The method is fully compatible with mainstream vectorization workflows and extends the capabilities of neural rendering for abstract vector representation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Deep Neural Architectures and Foundation Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

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.

Problem

• Generating smooth long paths for neural image abstraction
• Balancing fidelity and simplicity with parametric control
• Enabling stylization control in geometric and image spaces

Innovation

• Integrates smoothing B-splines into DiffVG pipeline
• Uses derivative-based costs for fidelity-simplicity tradeoffs
• Enables geometric and image space stylization control