This was a project created to experiment with machine learning and GAN image processing.
Unfortunately I don't have an nvidia gpu and AMDs support for machine learning is next to none.
What I did was train the model with images of myself and with the help of a discriminator running on my cpu. I had to utilize checkpoints since training a neural network on a CPU alone does take some time
Here are some of the best results that I have gotten from my neural net
Once tensorflow and AMD start supporting their 6800xt lineup i will definately be revisiting this project and experimenting with other types of machine learning branches and push the limits of what can be done
-
TENSORFLOW
-
KERAS
and create 3 directories in the root of the project
-
training_dataset/images -- you will put all of your images in this directory
-
training_model/checkpoint -- this will be your checkpoint directory
-
created/images/ -- this will be where all of the gererated images end up after each epoch cycle
if you are struggling to the the program to run, take a look at the official documentation here https://www.tensorflow.org/tutorials/generative/dcgan