Compare the value of DatasetMode enum, instead of Enum itself directly

[ankuPRK]

Hello, thanks so much for your contribution Marigold!

While working with the datasets, I found that the code was never going inside this if condition:

Marigold/src/dataset/base_normals_dataset.py

Line 96 in b1dffaa

rasters = self._training_preprocess(rasters)

Turns out, the DatasetMode enum defined at src/dataset/init.py which is used in train scripts comes from depth dataset:

Marigold/src/dataset/__init__.py

Line 37 in b1dffaa

DatasetMode,

Which gets compared to the DatasetMode enum defined at src/dataset/base_normals_dataset.py

Marigold/src/dataset/base_normals_dataset.py

Line 44 in b1dffaa

class DatasetMode(Enum):

This evaluates to False. Example:

>>> from src.dataset import DatasetMode as DatasetModeBase
>>> from src.dataset.base_depth_dataset import DatasetMode as DatasetModeDepth
>>> from src.dataset.base_normals_dataset import DatasetMode as DatasetModeNormals
>>> from src.dataset.base_iid_dataset import DatasetMode as DatasetModeIID
>>> DatasetModeBase.TRAIN == DatasetModeDepth.TRAIN
True
>>> DatasetModeBase.TRAIN == DatasetModeNormals.TRAIN
False
>>> DatasetModeBase.TRAIN == DatasetModeIID.TRAIN
False

It has a very simple fix: We should compare enum value instead:

>>> DatasetModeBase.TRAIN.value == DatasetModeDepth.TRAIN.value
True
>>> DatasetModeBase.TRAIN.value == DatasetModeNormals.TRAIN.value
True
>>> DatasetModeBase.TRAIN.value == DatasetModeIID.TRAIN.value
True
>>> DatasetModeBase.EVAL.value == DatasetModeIID.TRAIN.value
False

[markkua]

Thanks for catching this issue. I've created a fix at #168