//=time() ?>
Tried redoing this to check my progress. With the PhD taking up most of my time I don't get to practice as much as I'd like but still glad to see some improvement. #くりあファイル
Decided to retrain my GAN with the new pytorch version since apparently they decided to disable tf32 by default due to bugs, so far doesn't really stand out as being too different from previous runs
Further progress, fixed a bug in correcting face orientation when building the training dataset, which ought to improve average quality further
Much slower but maybe this won't get murdered by compression
Anime GAN project continues, with a more carefully curated dataset and various architectural adjustments
Still at it with the gan, changed the architecture a lot
Most recent GAN output, model is progressing very slowly at this point.
Results on latest #GAN! Main change compared to before was to use 1024 image batches throughout training with a wider network. A bit over a day and a half on 2 3090s.