SketchyCOCO: Image Generation from Freehand Scene Sketches

SketchyCOCO: Image Generation from Freehand Scene Sketches

We introduce the first method for automatic image generation from scene-level freehand sketches. Our model allows for controllable image generation by specifying thesynthesis goal via freehand sketches. The key contribution is an attribute vector bridged Generative Adversarial Network called EdgeGAN, which supports high visual-quality object-level image content generation without using freehand sketches as training data. We have built a largescale composite dataset called SketchyCOCO to support and evaluate the solution. We validate our approach on the tasks of both object-level and scene-level image generation on SketchyCOCO. Through quantitative, qualitative results, human evaluation and ablation studies, we demonstrate the method’s capacity to generate realistic complex scene-level images from various freehand sketches.

SketchyCOCO Dataset

In summary, we collect 20198(18869+1329) triplets of {foreground sketch, foreground image, foreground edge map} examples covering 14 classes, 27683(22171+5512) pairs of {background sketch, background image} examples covering 3 classes, 14081(11265+2816) pairs of {foreground image&background sketch, scene image} examples, 14081(11265+2816) pairs of {scene sketch, scene image} examples, and the segmentation ground truth for 14081(11265+2816) scene sketches.