SketchyCOCO: Image Generation from Freehand Scene Sketches

TL;DR

This paper presents EdgeGAN, a novel attribute vector bridged GAN for automatic, controllable scene-level image generation from freehand sketches, supported by a new large-scale dataset SketchyCOCO.

Contribution

The introduction of EdgeGAN enables high-quality scene and object image synthesis from sketches without training data containing sketches.

Findings

EdgeGAN produces realistic scene images from sketches.

The method outperforms existing approaches in quality and controllability.

Human evaluations favor the generated images' realism.

Abstract

We introduce the first method for automatic image generation from scene-level freehand sketches. Our model allows for controllable image generation by specifying the synthesis 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 large-scale 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.