Troubleshooting LoadAnnotations Error with Channel Masks

[erdikilic]

I have three channel masks, each representing one of three classes. The masks have binary values of 0 and 1 for the background and target. Unfortunately, there are some conflicting areas in the masks that prevent me from using multi-class labeling. I am currently using the LoadAnnotations function to load the mask PNGs, but I am encountering an error. How can I resolve this issue?

Traceback (most recent call last):
File "/content/mmsegmentation/tools/train.py", line 106, in
main()
File "/content/mmsegmentation/tools/train.py", line 102, in main
runner.train()
File "/usr/local/lib/python3.8/dist-packages/mmengine/runner/runner.py", line 1686, in train
model = self.train_loop.run() # type: ignore
File "/usr/local/lib/python3.8/dist-packages/mmengine/runner/loops.py", line 263, in run
data_batch = next(self.dataloader_iterator)
File "/usr/local/lib/python3.8/dist-packages/mmengine/runner/loops.py", line 158, in next
data = next(self._iterator)
File "/usr/local/lib/python3.8/dist-packages/torch/utils/data/dataloader.py", line 628, in next
data = self._next_data()
File "/usr/local/lib/python3.8/dist-packages/torch/utils/data/dataloader.py", line 1333, in _next_data
return self._process_data(data)
File "/usr/local/lib/python3.8/dist-packages/torch/utils/data/dataloader.py", line 1359, in _process_data
data.reraise()
File "/usr/local/lib/python3.8/dist-packages/torch/_utils.py", line 543, in reraise
raise exception
AssertionError: Caught AssertionError in DataLoader worker process 0.
Original Traceback (most recent call last):
File "/usr/local/lib/python3.8/dist-packages/torch/utils/data/_utils/worker.py", line 302, in _worker_loop
data = fetcher.fetch(index)
File "/usr/local/lib/python3.8/dist-packages/torch/utils/data/_utils/fetch.py", line 58, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/usr/local/lib/python3.8/dist-packages/torch/utils/data/_utils/fetch.py", line 58, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/usr/local/lib/python3.8/dist-packages/mmengine/dataset/base_dataset.py", line 408, in getitem
data = self.prepare_data(idx)
File "/usr/local/lib/python3.8/dist-packages/mmengine/dataset/base_dataset.py", line 789, in prepare_data
return self.pipeline(data_info)
File "/usr/local/lib/python3.8/dist-packages/mmengine/dataset/base_dataset.py", line 58, in call
data = t(data)
File "/usr/local/lib/python3.8/dist-packages/mmcv/transforms/base.py", line 12, in call
return self.transform(results)
File "/content/mmsegmentation/mmseg/datasets/transforms/formatting.py", line 74, in transform
data_sample.gt_sem_seg = PixelData(**gt_sem_seg_data)
File "/usr/local/lib/python3.8/dist-packages/mmengine/structures/base_data_element.py", line 219, in init
self.set_data(kwargs)
File "/usr/local/lib/python3.8/dist-packages/mmengine/structures/base_data_element.py", line 249, in set_data
setattr(self, k, v)
File "/usr/local/lib/python3.8/dist-packages/mmengine/structures/pixel_data.py", line 78, in setattr
assert value.ndim in [
AssertionError: The dim of value must be 2 or 3, but got 4

[MeowZheng]

I think you can convert the segmentation masks of your dataset into single channel mask.
here is an example:

mmsegmentation/tools/convert_datasets/isaid.py

Line 37 in b2fdae7

def iSAID_convert_from_color(arr_3d, palette=iSAID_invert_palette):

[erdikilic]

Actually, single channel mask solution is not my case. As I mentioned before, some classes have joint masks. For example, a pixel target can be [0,1,1] as a result of three labels. But, I just wonder that, can I use LoadBiomedicalAnnotation to load multichannel masks for a PNG image with LoadImage?

[MeowZheng]

I fix it in #2739
LoadBiomedicalAnnotation is for 3D biomedical data, I think it is improper for your project