Using GANs to Generate Realistic Art
We are using GANs to generate art. We will be attempting to create abstract art, as well as portrait and landscape art.
Updates
Introduction: We are generating art. We will begin by working with the abstract art dataset, but we hope to generate art based upon the landscape and portrait datasets as well. This is an unsupervised learning problem.
Challenges: We've run into a few difficult bugs in implementing a simple GAN network. Our biggest challenge that we are working on currently is changing our generated images from black and white to color by creating 3 channels.
Insights: We don't have any concrete results from our model just yet. Although we expected training GANs to be difficult, it has surpassed our expectations.
Plan: Currently we are using dense layers in our generator and discriminator networks, but we will change these to convolution layers. We are also currently only using 1 channel, so our images are coming out black and white, so we will be changing this to have three channels for RGB.
https://github.com/nhyland46/GAN_abstract
Log in or sign up for Devpost to join the conversation.
Leave feedback in the comments!
Log in or sign up for Devpost to join the conversation.
Log in or sign up for Devpost to join the conversation.