ClipGen: A Deep Generative Model for Clipart Vectorization and Synthesis

In this paper, we present a novel deep learning-based approach for automatically vectorizing and synthesizing clipart of man-made objects. Given a raster clipart image and it's corresponding object category (e.g., airplane), our method sequentially generates new layers. Each layer is composed of a new closed path filled with a single color. The final result is obtained by compositing all layers together into a vector clipart that falls into the target category. At the core of our approach is an iterative generative model that decides (i) whether to keep synthesizing a new layer and (ii) the geometry and appearance of the new layer. We formulate a joint loss function for training our generative model, including shape similarity, symmetry, local curve smoothness losses, and a vector graphics rendering accuracy loss to synthesize recognizable clipart. We also introduce a collection of man-made object clipart, ClipNet, composed of layers of a closed path, and we design two preprocessing tasks to clean up and enrich the original raw clipart. To validate our approach, we perform several validation studies and demonstrate the ability to vectorize and synthesize various clipart categories. We envision that our generative model can facilitate efficient and intuitive clipart design for novice users and graphic designers.