CoGS: Controllable Generation and Search from Sketch and Style
This addresses the need for controllable image generation for users in creative or design applications, though it is incremental as it builds on existing transformer and VQGAN techniques.
The authors tackled the problem of generating images with decoupled control over structure and appearance by introducing CoGS, a method that uses sketches and style images to synthesize diverse outputs, achieving a wide range of semantic content and styles on a new dataset of 125 object classes.
We present CoGS, a novel method for the style-conditioned, sketch-driven synthesis of images. CoGS enables exploration of diverse appearance possibilities for a given sketched object, enabling decoupled control over the structure and the appearance of the output. Coarse-grained control over object structure and appearance are enabled via an input sketch and an exemplar "style" conditioning image to a transformer-based sketch and style encoder to generate a discrete codebook representation. We map the codebook representation into a metric space, enabling fine-grained control over selection and interpolation between multiple synthesis options before generating the image via a vector quantized GAN (VQGAN) decoder. Our framework thereby unifies search and synthesis tasks, in that a sketch and style pair may be used to run an initial synthesis which may be refined via combination with similar results in a search corpus to produce an image more closely matching the user's intent. We show that our model, trained on the 125 object classes of our newly created Pseudosketches dataset, is capable of producing a diverse gamut of semantic content and appearance styles.