Multi-view Convolutional Neural Networks for 3D Shape Recognition

@inproceedings{su2015multi,
  title={Multi-view convolutional neural networks for 3d shape recognition},
  author={Su, Hang and Maji, Subhransu and Kalogerakis, Evangelos and Learned-Miller, Erik},
  booktitle={Proceedings of the IEEE international conference on computer vision},
  pages={945--953},
  year={2015}
}

Architecture #

Each view goes through the same CNN
The results are then pooled
After a second CNN , a class is predicted
Special: The first CNN can be pretrained, using powerful image processing CNNs

Results #

Figure 1: Performance of MVCNN is better than some nets working directly on the 3D data itself.

Considerations #

Maybe in real world usage / test scenario, provided views are not carefully selected as for training