Simple 3D GAN to demonstrate interpolation

What

This jupyter notebook provides a sample interpolation using the GAN from https://github.com/rp2707/coms4995-project

It trains on the Shapenet dataset particularly two models, a chair and a airplane (~17600 3D volumetric objects). Goal is to interpolate between two noises, one giving a artificially generated chair and another one giving a airplane.

The interesting part is the vector algebra you could use to create mixtures of Objects like: Chair/2 + Airplane/2 = mix of both

How

Requirements:

• Python >= 3.
• Tensorflow
• Scipy
• Jupyter notebook
• numpy
• Matplotlib

the shellscript downloads the pretrained models (no training needed) and the 3dshapenet data. After downloading you can check the jupyter notebook

Literature

• http://3dgan.csail.mit.edu/
• https://github.com/rp2707/coms4995-project/tree/master/multicategory