Add MoNuSAC Dataset

[anwai98]

(WIP) Haven't been tested, as the generate_labeled_array depends on the MoNuSeg PR (will test and put to review once this works)

[anwai98]

There's an error thrown by an input image (TCGA-5P-A9K0-01Z-00-DX1_3.tif) of shape (185, 497, 4), while the dataloader requests for a patch shape of (512, 512), stating:

Traceback (most recent call last):
  File "/home/nimanwai/torch-em/scripts/datasets/check_monusac.py", line 32, in <module>
    check_monusac()
  File "/home/nimanwai/torch-em/scripts/datasets/check_monusac.py", line 18, in check_monusac
    check_loader(train_loader, 8, instance_labels=True, rgb=True, plt=True, save_path="./monusac_train.png")
  File "/home/nimanwai/torch-em/torch_em/util/debug.py", line 127, in check_loader
    _check_plt(loader, n_samples, instance_labels, save_path=save_path)
  File "/home/nimanwai/torch-em/torch_em/util/debug.py", line 17, in _check_plt
    for ii, (x, y) in enumerate(loader):
  File "/scratch/usr/nimanwai/mambaforge/envs/sam/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 634, in __next__
    data = self._next_data()
  File "/scratch/usr/nimanwai/mambaforge/envs/sam/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 678, in _next_data
    data = self._dataset_fetcher.fetch(index)  # may raise StopIteration
  File "/scratch/usr/nimanwai/mambaforge/envs/sam/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 51, in fetch
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/scratch/usr/nimanwai/mambaforge/envs/sam/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 51, in <listcomp>
    data = [self.dataset[idx] for idx in possibly_batched_index]
  File "/home/nimanwai/torch-em/torch_em/data/image_collection_dataset.py", line 167, in __getitem__
    raw, labels = self._get_sample(index)
  File "/home/nimanwai/torch-em/torch_em/data/image_collection_dataset.py", line 134, in _get_sample
    raw, label = self._ensure_patch_shape(raw, label, have_raw_channels, have_label_channels, channel_first)
  File "/home/nimanwai/torch-em/torch_em/data/image_collection_dataset.py", line 106, in _ensure_patch_shape
    raise NotImplementedError("Padding is not implemented for data with channels")
NotImplementedError: Padding is not implemented for data with channels

whereas, if I request for a patch shape of (128. 128), it works just fine.

What do you think about this @constantinpape? (I think the issue is that the paddding implementation for multi-channel inputs is missing)

[anwai98]

@constantinpape The padding for images with raw and label channels has been taken care of now. Let me know how this looks now.

[constantinpape]

Besides the comments I left in the code: this data is saved as RGBA. That doesn't make sense, the alpha channel (A) does not contain any information and we should get rid of it.

[anwai98]

Besides the comments I left in the code: this data is saved as RGBA. That doesn't make sense, the alpha channel (A) does not contain any information and we should get rid of it.

Okay. So for this, what would you recommend? (to manually remove the alpha channels from the tif images? ~ using a functionality on top)

(UPDATE: I implemented a funcitonality which takes care of this in the recent commit)

[constantinpape]

Looks good now, you should just change how the RGBA data is converted.