Making Robots Draw A Vivid Portrait In Two Minutes

  • 2020-05-12 03:02:24
  • Fei Gao, Jingjie Zhu, Zeyuan Yu, Peng Li, Tao Wang
  • 47

Abstract

Significant progress has been made with artistic robots. However, existingrobots fail to produce high-quality portraits in a short time. In this work, wepresent a drawing robot, which can automatically transfer a facial picture to avivid portrait, and then draw it on paper within two minutes averagely. At theheart of our system is a novel portrait synthesis algorithm based on deeplearning. Innovatively, we employ a self-consistency loss, which makes thealgorithm capable of generating continuous and smooth brush-strokes. Besides,we propose a componential-sparsity constraint to reduce the number ofbrush-strokes over insignificant areas. We also implement a local sketchsynthesis algorithm, and several pre- and post-processing techniques to dealwith the background and details. The portrait produced by our algorithmsuccessfully captures individual characteristics by using a sparse set ofcontinuous brush-strokes. Finally, the portrait is converted to a sequence oftrajectories and reproduced by a 3-degree-of-freedom robotic arm. The wholeportrait drawing robotic system is named AiSketcher. Extensive experiments showthat AiSketcher can produce considerably high-quality sketches for a wide rangeof pictures, including faces in-the-wild and universal images of arbitrarycontent. To our best knowledge, AiSketcher is the first portrait drawing robotthat uses deep learning techniques. AiSketcher has attended a quite number ofexhibitions and shown remarkable performance under diverse circumstances.

 

Quick Read (beta)

loading the full paper ...