Artistic Explorations into GANs
and Neural Style Transfer
Print & Video
An exploration into how generative deep learning works and what kind of experimentations can be done using machine learning. My main focus was looking at the results visually hence I approached all outcomes with an eye for aesthetics and experimentation. Since I had a very broad starting point I narrowed my project down to using two ML implementations - GAN (Generative Adversarial Network) and Neural Style Transfer using tf.Keras
Prior to experimenting with GAN and Style Transfer I also did a small experiment with T-sne using my own photographs from the past 5 years. T-SNE is also pre-trained on image net which is a library of millions of images from the internet - it’s main fucntion is to cluster or group things together based on some parameters of similarity.
It was interesting the way it decided to map them in an order - it has arranged images according to landscapes, pictures of groups of people, portraits, pictures taken during nighttime versus daytime, pictures of clouds and textures.
Read full paper here