eduzhai > Applied Sciences > Engineering >

BézierSketch A generative model for scalable vector sketches

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 19 pages

Abstract: The study of neural generative models of human sketches is a fascinatingcontemporary modeling problem due to the links between sketch image generationand the human drawing process. The landmark SketchRNN provided breakthrough bysequentially generating sketches as a sequence of waypoints. However this leadsto low-resolution image generation, and failure to model long sketches. In thispaper we present BézierSketch, a novel generative model for fully vectorsketches that are automatically scalable and high-resolution. To this end, wefirst introduce a novel inverse graphics approach to stroke embedding thattrains an encoder to embed each stroke to its best fit Bézier curve. Thisenables us to treat sketches as short sequences of paramaterized strokes andthus train a recurrent sketch generator with greater capacity for longersketches, while producing scalable high-resolution results. We reportqualitative and quantitative results on the Quick, Draw! benchmark.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×