Will Hoover

Week of 6/23 – 6/30/2024

This week I decided to investigate the prompts being used for training and how that worked for our StyleGAN-T model. When I had the model read out what the label was, it returned “000000/img000000.png” which would be the folder and the name of the image, not the prompt for that image. So, after adjusting the script I had it dig correctly into the database.json file returning the correct prompts, but this led to another problem. Now I am getting an Out of Memory (OOM) error stating my graphics card doesn’t have enough memory to train the model, but I am getting this error even with a batch size of 1. This is leading me to believe that I’m either not feeding the labels correctly or the training images need to go from 256*256 down to 128*128.

I updated the repo debug_generator with the code I have been working on this week.

Thanks for checking in

-Will Hoover