No description, website, or topics provided.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
2 to 8.png
CelebA interpolation.png
DCGAN-FACES_AWS.ipynb
DCGAN-MNIST.ipynb
README.md
faces.gif
faces_original.png
mnist.gif
mnist_original.png
right to left.gif
woman to man.gif

README.md

DCGAN Image Generation

What would it be like to design a chair 20 years from now? Do we still have to bear with the complexity of the design software and spend the entire week clicking here and there? Possibly. Or, maybe we can let the computer handle the heavy lifting to generate designs that are good and we pick the ones we like!

Researchers are working on bringing this Sci-Fi scene to reality and they named it Generative Design. In this project I took a peek into this lovely technology by having implemented DCGAN, which is a novel framework of ConvNets to generate visually realistic images (both B&W and color).

Original DCGAN paper: https://arxiv.org/pdf/1511.06434v2.pdf

More about Generative Design: https://redshift.autodesk.com/what-is-generative-design-2/

Data Sources

MNIST: http://yann.lecun.com/exdb/mnist/
CelebA: http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html

MNIST

Original MNIST samples:

Generated MNIST samles:

Linear Interpolation from 2 to 8

CelebA

Original CelebA samples:

Generated CelebA samples:

Spherical Interpolation for generated faces