Project Description

In this project, our ultimate goal is to build a robust pipeline to vectorize scanned images of floor plans and/or other drawings for the processing tasks (editing, etc.). We approach this problem by dividing it to simpler tasks such as:

Here are some comments on each step of the pipeline

For vectorization we tried to use many algorithmic methods for example portrace, fidelity vs simplicity etc. There results are not satisfactory. Therefore we are developing an architecture based on Attention network with differentiable rendering loss

References

  1. Potrace: https://en.wikipedia.org/wiki/Potrace
  2. Favreau, Jean-Dominique and Lafarge, Florent and Bousseau, Adrien. Fidelity vs. Simplicity: a Global Approach to Line Drawing Vectorization. ACM Transactions on Graphics (SIGGRAPH Conference Proceedings), 2016, http://www-sop.inria.fr/reves/Basilic/2016/FLB16
  3. Chen, Wengling and Hays, James. SketchyGAN: Towards Diverse and Realistic Sketch to Image Synthesis. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018
  4. Kim, Byungsoo and Wang, Oliver and Oztireli, A. Cengiz and Gross, Markus. Semantic Segmentation for Line Drawing Vectorization Using Neural Networks. Computer Graphics Forum (Proc. Eurographics), 2018, volume 37, number 2, pages 329-338