A Learned Representation for Scalable Vector Graphics

Abstract

Dramatic advances in generative models have resulted in near photographic quality for artificially rendered faces, animals and other objects in the natural world. In spite of such advances, a higher level understanding of vision and imagery does not arise from exhaustively modeling an object, but instead identifying higher-level attributes that best summarize the aspects of an object. In this work we attempt to model the drawing process of fonts by building sequential generative models of vector graphics. This model has the benefit of providing a scale-invariant representation for imagery whose latent representation may be systematically manipulated and exploited to perform style propagation. We demonstrate these results on a large dataset of fonts and highlight how such a model captures the statistical dependencies and richness of this dataset. We envision that our model can find use as a tool for graphic designers to facilitate font design. * Work done as a member of the Google AI Residency Program (g. co/airesidency) Learned Vector Graphics RepresentationPixel Counterpart moveTo (15, 25) lineTo (-2, 0.3) cubicBezier

Differentiable vector graphics rasterization for editing and learning

“…Deep learning models have been proposed for vector graphics. Ha and Eck [2018] and Lopes et al [2019] trained generative recurrent models using a sketch database. Azadi et al [2018] and Yue et al [2019] generate new fonts given a few example styles.…”

“…The training took about 12 hours for each model. For more details of the architecture and training, see Appendix C. Thanks to our differentiable rasterization, we can train generative models with only image-based supervision, without requiring any supervision on the curves themselves like previous approaches [Ha and Eck 2018;Lopes et al 2019]. Neither do we need to train a neural network for approximating the rasterization process, which usually fixes and limits the output resolution and only works on a single type of vector primitive Nakano 2019;.…”

“…As such, applications presented above are limited to fixed topologies, determined either heuristically, or by preserving the topology of the input. It is possible to use a recurrent or reinforcement learning model to handle the discrete decisions [Ha and Eck 2018; Lopes et al 2019;Zhu et al 2018], while additionally benefiting from being able to compute gradients on continuous parameters.…”

ClipFlip : Multi‐view Clipart Design

“…Meanwhile, Lopes et al. [LHES19a] proposed a generative model for font synthesis with a deep learning framework handles vector format. However, we can not apply this method directly for our application because their model did not handle (i) the colour, (ii) shape category and (iii) the viewpoint information.…”

“…Moreover, this paper aims not to propose a novel deep learning architecture that directly handles vector clipart. Even though there exists relevant work [LHES19b] that handles vector format data using a deep generative model, however, their model did not consider (i) the colour, (ii) shape category and (iii) the viewpoint information. These pieces of information are vital to our application.…”