Deep Sketch-Based Modeling of Man-Made Shapes

Deep Sketch-Based Modeling of Man-Made Shapes

@article{smirnov2019deep,
  author    = {Dmitriy Smirnov and
               Mikhail Bessmeltsev and
               Justin Solomon},
  title     = {Deep Sketch-Based Modeling of Man-Made Shapes},
  journal   = {CoRR},
  volume    = {abs/1906.12337},
  year      = {2019},
  url       = {http://arxiv.org/abs/1906.12337},
  eprinttype = {arXiv},
  eprint    = {1906.12337},
  timestamp = {Thu, 16 Apr 2020 07:55:10 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-1906-12337.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}
  • Input 2d sketch
  • Ouput: generated 3d shape
  • Generate Coons Patch
    • Based on set of (per category) templates (bounding boxes)
  • Each category has
    • a ResNet encoder
    • fully connected layers (MLP ) to predict control points
  • Reconstruction loss:
    • Chamfer Distance
    • Normal alignment
    • Self-Collision penalization
    • patch flatnetss regulizer (flat patches preferred)
Calendar October 22, 2023