CanvasVAE: Learning to Generate Vector Graphic Documents

TL;DR

CanvasVAE introduces a generative model for vector graphic documents, capturing their complex structure and attributes, and provides a new dataset for training and evaluation in this domain.

Contribution

We propose CanvasVAE, a variational auto-encoder that models vector graphic documents with multi-modal attributes and structure, along with a new dataset of design templates.

Findings

CanvasVAE effectively generates vector graphic documents.

The model captures complex document structures and occluded elements.

Our dataset enables comprehensive training and evaluation.

Abstract

Vector graphic documents present visual elements in a resolution free, compact format and are often seen in creative applications. In this work, we attempt to learn a generative model of vector graphic documents. We define vector graphic documents by a multi-modal set of attributes associated to a canvas and a sequence of visual elements such as shapes, images, or texts, and train variational auto-encoders to learn the representation of the documents. We collect a new dataset of design templates from an online service that features complete document structure including occluded elements. In experiments, we show that our model, named CanvasVAE, constitutes a strong baseline for generative modeling of vector graphic documents.