A Neural Representation of Sketch Drawings

TL;DR

This paper introduces sketch-rnn, a recurrent neural network capable of generating coherent stroke-based sketches of common objects, trained on extensive human-drawn data, with methods for both conditional and unconditional sketch synthesis.

Contribution

The paper presents a novel RNN architecture for sketch generation, along with robust training techniques and a flexible framework for conditional and unconditional drawing synthesis.

Findings

Able to generate coherent sketches of various objects

Trained on thousands of human-drawn images across hundreds of classes

Provides a framework for flexible sketch generation

Abstract

We present sketch-rnn, a recurrent neural network (RNN) able to construct stroke-based drawings of common objects. The model is trained on thousands of crude human-drawn images representing hundreds of classes. We outline a framework for conditional and unconditional sketch generation, and describe new robust training methods for generating coherent sketch drawings in a vector format.