I worked with 3 other students on implementing a Generative Adversarial Network. GANs is a type of neural network architecture used for generating instances of training data. For instances, when images are fed into GANs, an algorithm can create or generate similar images never seen before. This technique is used in DeepFakes. It is a very ethically sensitive technology, and we considered all of it.

As a demonstration, we compared two types of GANs viz  Fully Connected(FC) and Deep Convolutional(DC) GANs on 4 datasets, MNIST, FASHION-MNIST, CIFAR10 and CIFAR100, using the Frechet Inception Distance score.

The results are outstanding. More details of this can be found in the poster below.


The source code can be found here https://github.com/Chuukwudi/FC-GAN-vs-DC-GAN-comparison-on-MNIST-MNIST_Fashion-Cifar10-and-Cifar100-Datasets and can be directly run on google colab.

Skip to content
%d bloggers like this: