Motivation  

This work presents a classification-based approach for Next View Selection. We propose a two-stage CNN network to simultaneously obtain next view and reconstruct based on the input images. In this process, we propose a loss function which is end-to-end trainable. The proposed model is unique because we have a classifier layer followed by a reconstructor network. Our model outperforms both state of the art 3D shape reconstruction methods and next view selection methods.

Please refer to the paper for more details: https://arxiv.org/abs/2012.01743