In this recipe, we're going to cover a few basic techniques for producing 3D models with GANs and we'll see how we can simplify these architectures for learning.
There are three papers that you'll need to review to understand this recipe:
- Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling: http://3dgan.csail.mit.edu/papers/3dgan_nips.pdf.
- Learning Shape Priors for Single-View 3D Completion and Reconstruction: http://shapehd.csail.mit.edu/papers/shapehd_eccv.pdf.
- Interactive 3D Modeling with a Generative Adversarial Network: https://arxiv.org/pdf/1706.05170.pdf.
There are two major steps in the process to go from 2D images to 3D voxelized models—encoding and 3D convolutions.
In each section, we'll cover the basics of the concepts that you'll use throughout this chapter.