Will Hoover

Week of 5/26 – 6/2/2024

This week was spent working on the troubleshooting the model, first on the list was to get the model to recognize the labels that need to be read and applied to the model. The problem turned out to be a pathing issue, which happens to be one of my weakest areas. I was able to fix this by adding the folder it was in within to the name of the file that I fed into my extract zip function. Once the training was properly reading the labels and applying them to training, as we expected I needed to start training over again. This has led to a few problems, before I got the labels to properly read, the model trained fine for about 300 ticks with no problems, and it was able to start from a trained .pkl file. Since the labels have been reading and being fed in, I cannot get my network to train past epoch 20. No matter what changes I make, or ways I go about things I cannot get it to train past 20 epochs and I cannot get it to train from a saved .pkl file as before.  

I am going to get with my team on Monday 6/3/2024 to see if another set of eyes would be helpful in finding the problem. Once the training is fixed and this is able to train and load from a .pkl, getting the 400 ticks I want before moving on to the FFHQ dataset shouldn’t take more than a few week.

Will Hoover