3D Vision

Multi-View Convolutional Neural Networks

1. Take different views of the same object
2. Use a ConvNet to extract image features
3. Max pooling across all views
4. A second ConvNet producing shape descriptors
   

3D Shape Representations