Questions et al.
3D-R2N2: A Unified Approach for Single and Multi-View 3D Object Reconstruction

Send your answers to zhouh@informatik.uni-freiburg.de before June 1st, 2016, 10am

1) What's the difference between recurrent neural networks and feedforward neural networks? Why is RNN preferred in this paper? (2-3 sentences)

2) In this paper a new architecture 3D-Convolutional LSTM is proposed. How is 3D-LSTM extended from standard LSTM and why is the extension beneficial? (2-3 sentences)

3) How is the 3D shape represented in this paper? How can the network output such 3D representation and what loss is used? (2-3 sentences)