COMS 4995 Deep Learning Project, Spring 2018 - Gp6
We implemented a 3D-VAE-GAN architecture for 3D object generation and reconstruction.
Challenges of this project mainly come from two perspectives:
- The model needs to be able to generate 3D objects that are realistic as well as varied from the objects it has seen in the training phase;
- Instead of arbitrarily generating some meaningless 3D objects, the model should be capable of taking certain observations as input and generate corresponding 3D objects.
The goal of this project is to solve above challenges by creating a network that can not only synthesize 3D objects but also learn some latent representations from 2D images and reconstruct 3D objects base on them.