# Learning to Infer Graphics Programs from Hand-Drawn Images (@ NeurIPS 2018)

### Kevin Ellis, Daniel Ritchie, Armando Solar-Lezama, Josh Tenenbaum

All right, here comes the core idea, simple and neat. Suppose you have hand-drawn images. One way to extract "what" is being drawn, as opposed to just what pixels are on and off, is to try to find a small program that generates that image. This program will be written in a DSL that has primitive operations like "draw a circle at $(x, y)$ with radius $r$", or draw an arrow from $(x_1, y_1)$ to $(x_2, y_2)$. Moreover, it can have loops and conditionals, that can capture, for instance, that the drawing is a grid of squares.