SketchGNN: Semantic Sketch Segmentation with Graph Neural Networks

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