Project information
- Category: Deep Learning
- Purpose: Durham University
- Assignment Grade: 84%
- Project date: May, 2021
- Github URL: https://colab.research.google.com/drive/1jMFWkJyYAfdTIfk_HppUWeJenE7doFCK?usp=sharing
DCGAN Pegasus Coursework
I created a DCGAN to combine images from the CIFAR-10 and STL-10 datasets to generate images of pegasi. This involved testing numerous methods of combining images of bird, planes, and horses, such that the resultant image most closely resembled the mythical creature. Furthermore, additional experimentation with greyscale versions and image editing was conducting as white pegasi were preferred in the specification. The final solution involved loading the plane and horse datasets in weighted, alternating steps, such that a combination of the two datasets was created.
The best batch of pegasi created:
The best singular pegasus created:
My Report for this project can be found below:
Open on desktop to view PDF