- Get data from globus
- Differentiate train and test
- Split into better sizes
- Without torchrun, make setup and destroy function and make sure we can use multiple gpus on idun
- Hyperparams stored in json file, try to load
- Able to initialize trainer
- save and load from checkpoint file
- setup full end to end training
- Encoder
- Decoder
- Elbo loss that works
- forward pass, sampling and recon works
- Pictures before
- Pictures after
- Imgs with found anomlies
- TP, FP and so on