SketchGNN: Semantic Sketch Segmentation with Graph Neural Networks

TL;DR

SketchGNN is a graph neural network that effectively segments and labels freehand sketches by modeling strokes as graphs, significantly outperforming previous methods in accuracy with fewer parameters.

Contribution

The paper introduces SketchGNN, a novel graph neural network architecture that improves semantic sketch segmentation by leveraging stroke structure and multi-level feature extraction.

Findings

11.2% improvement in pixel-based accuracy

18.2% improvement in component-based accuracy

Fewer parameters than existing methods

Abstract

We introduce SketchGNN, a convolutional graph neural network for semantic segmentation and labeling of freehand vector sketches. We treat an input stroke-based sketch as a graph, with nodes representing the sampled points along input strokes and edges encoding the stroke structure information. To predict the per-node labels, our SketchGNN uses graph convolution and a static-dynamic branching network architecture to extract the features at three levels, i.e., point-level, stroke-level, and sketch-level. SketchGNN significantly improves the accuracy of the state-of-the-art methods for semantic sketch segmentation (by 11.2% in the pixel-based metric and 18.2% in the component-based metric over a large-scale challenging SPG dataset) and has magnitudes fewer parameters than both image-based and sequence-based methods.