published on December 9th, 2019
One of the projects I want to try in my ultralearning datascience project is to create a neural network model that can convert images to SVG files.
I looked around for existing literature, models and programs that try this.
The current state-of-the-art program seems to be VectorMagic. The technology behind VectorMagic is not clear, however it was created in 2007 and one of the founders was previously working at the Stanford AI lab working on path planning for autonomous vehicles, so it seems unlikely that it relies directly on deep learning architectures
There is the pix2code paper and corresponding github repo by Tony Beltramellif of uizard which does something similar trying to create UI code for iOS and Android applications from images. There is also a follow-up project called code2pix trying to improve on the initial results.
On this question on quora Ian Goodfellow points out the pointer networks paper which is using neural networks to predict the convex hull of a set of points.