Robin Cole

@adamjstewart using ```python TARGET_SIZE = (282, 282) ... resize = K.Resize(size=TARGET_SIZE) extra_args = { DataKey.MASK: {'resample': Resample.NEAREST, 'align_corners': None} } if mode == 'time-series': self.aug = K.AugmentationSequential( K.VideoSequential(resize, normalizer), data_keys=None,...

I've iterated though a few images and they just don't appear to match the mask: ```python datamodule = CopernicusBenchBiomassS3DataModule( root="/data/", batch_size=4, num_workers=4, ) datamodule.setup(stage="fit") train_loader = datamodule.train_dataloader() batch = next(iter(train_loader))...

@wangyi111 can you share your code for the RGB plotting in the paper? Adding a utility for that, examples below:

I didn't notice the base class plot - these (using the base class plot) look decent so my hacky plot approach in a notebook must have an error in it

@adamjstewart codecov still short

New error now I am testing `time-series`: ``` RuntimeError: stack expects each tensor to be equal size, but got [21, 8, 8] at entry 0 and [21, 8, 10] at...

@adamjstewart all green

IDK inspecting another batch, these just don't look right. @wangyi111 WDYT?

I have managed to train a model, so perhaps I am just unused to viewing the S3 images. However given these are large areas of forest, perhaps that is possible...

@adamjstewart upgraded to `kornia-0.8.2` and the image/mask still look unregistered. I don't think this is a code issue, perhaps the images themselves are not well registered. Suggest a follow up...