Interactive Sketch & Fill: Multiclass Sketch-to-Image Translation
Arnab Ghosh1      Richard Zhang2      Puneet K. Dokania1
Oliver Wang2     Alexei A. Efros2,3     Philip H.S. Torr1     Eli Shechtman2

1 University of Oxford      2 Adobe Research      3 UC Berkeley

In ICCV 2019

We provide a sketch recommendation system which helps the user to draw and interactively generate realistic images.



We propose an interactive GAN-based sketch-to-image translation method that helps novice users easily create images of simple objects. The user starts with a sparse sketch and a desired object category, and the network then recommends its plausible completion(s) and shows a corresponding synthesized image. This enables a feedback loop, where the user can edit the sketch based on the network's recommendations, while the network is able to better synthesize the image that the user might have in mind. In order to use a single model for a wide array of object classes, we introduce a gating-based approach for class conditioning, which allows us to generate distinct classes without feature mixing, from a single generator network.

Method overview

The user makes an input stroke, and the shape generator network gives multiple shape completions based on the selected class. The appearance generator takes one of the shape completions and generates a realistic image from it.

Paper and Supplementary Material

A.Ghosh, R. Zhang, P. Dokania, O. Wang, A. Efros, P. Torr, E. Shechtman
Interactive Sketch & Fill: Multiclass Sketch-to-Image Translation
In ICCV, 2019.
(hosted on ArXiv)



We thank Harkirat Singh Behl, Noa Fish, Taesung Park, Rahul Arora for their helpful comments and feedback on the paper. AG, PKD, and PHST are supported by the ERC grant ERC-2012-AdG, EPSRC grant Seebibyte EP/M013774/1, EPSRC/MURI grant EP/N019474/1 and would also like to acknowledge the Royal Academy of Engineering and FiveAI. Part of the work was done while AG was an intern at Adobe.