Pixel2Mesh: Generating 3D Mesh Models from Single RGB Images

@inproceedings{wang2018pixel2mesh,
  title={Pixel2mesh: Generating 3d mesh models from single rgb images},
  author={Wang, Nanyang and Zhang, Yinda and Li, Zhuwen and Fu, Yanwei and Liu,
          Wei and Jiang, Yu-Gang},
  booktitle={Proceedings of the European conference on computer vision (ECCV)},
  pages={52--67},
  year={2018}
}

• Generate 3D Mesh from single RGB images
• Use a template shape that will be deformed in the process
• Limited by the topology of the template mesh

Method #

1. Take a template mesh (e.g. an Ellipsoid mesh)
2. Input image -> 2D CNN
3. Graph Neural Network to predict vertex displacements
4. Fixed upsampling (mesh subdivision) steps

• Chamfer Distance based Loss Function
• Regularization to prevent spikes or self intersection