Harsha

explicit repeat of dataset before calling fit method

2

For reasons described [here](https://github.com/neuronets/nobrainer/issues/313) and discussed elsewhere, I suggest repeat should be called before fit and not before performing any map operations in the Dataset class, which is redundant.

Retraining kwyk model (bayesian neural network) on kwyk dataset using nobrainer API

1

- [x] Try running the model until `fit` and document any issues - [ ] Run with `predict` and see how the API needs to be adapted. This is because...

Referring to https://github.com/neuronets/nobrainer/blob/902a124f34b03f18212c25754a2f2d4ea9b35760/nobrainer/tfrecord.py#L197-L203 and https://github.com/neuronets/nobrainer/blob/902a124f34b03f18212c25754a2f2d4ea9b35760/nobrainer/tfrecord.py#L230-L235 why do we serialize the shapes individually when the array could be serialized? Check if this has to do with https://github.com/neuronets/nobrainer/blob/902a124f34b03f18212c25754a2f2d4ea9b35760/nobrainer/tfrecord.py#L146-L149

https://github.com/neuronets/nobrainer/blob/902a124f34b03f18212c25754a2f2d4ea9b35760/nobrainer/tfrecord.py#L233 While this key is not being used it makes sense to rename it to `label/shape` for consistency

Fix PGAN issues

7

From @satra : - There is an issue with loading checkpoints (in multi-GPU case). Notes (03/22/2024) - could not run the example on dgx100, so moving it to CPU for...

https://github.com/neuronets/nobrainer/blob/902a124f34b03f18212c25754a2f2d4ea9b35760/nobrainer/processing/segmentation.py#L41-L44 I forgot to delete this when unrolling the changes from https://github.com/neuronets/nobrainer/issues/303#issuecomment-2015671569

Add test_unetr() in model_tests.py

1

related to https://github.com/neuronets/nobrainer/issues/302

The current test suite is limited to `binary_crossentropy`. Need to expand the range of tests.

What is the correct way to specify a loss? This will be relevant for streamlining the process of defining new losses. `losses.dice` as seen in https://github.com/neuronets/nobrainer/blob/976691d685824fd4bba836498abea4184cffd798/nobrainer/processing/segmentation.py#L54 or `losses.Dice()` as seen...

Rewrite metrics module with model subclassing

1

`What would you like changed/added and why?` Rewrite functions in metrics.py with model subclassing. See [here](https://keras.io/guides/training_with_built_in_methods/#custom-metrics). The drawback of the current approach is that metrics have to be kept track...