Camvid multiclass segmentation to single class implementation

[Shawn94]

I`ve tried to use a camvid example(ipynb file), wherefrom multiclass you can choose one preferred class and do training with it. As I understood mask images are in RGB shape, therefore have 3 channels. However, in my case, I have a single class dataset, where mask images are in black and white. If I read the image using cv2 and giving cv2.IMREAD_GRAYSCALE parameter I have error in training. Image shape (512,640,3), mask shape (512,640). If I try to expand dims of the mask to 1 or read masks as RGB, like original images(3 channels), training works well, but dice_loss gets minus numbers and iou score goes beyond 100%.
The error where I load mask as grayscale:
ValueError: Caught ValueError in DataLoader worker process 0.
Original Traceback (most recent call last):

File "/home/goback_main/anaconda3/envs/pl/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 287, in _worker_loop
data` = fetcher.fetch(index)
File "/home/goback_main/anaconda3/envs/pl/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 44, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/goback_main/anaconda3/envs/pl/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 44, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "", line 62, in getitem
sample = self.preprocessing(image=image, mask=mask)
File "/home/goback_main/anaconda3/envs/pl/lib/python3.8/site-packages/albumentations/core/composition.py", line 210, in call
data = t(force_apply=force_apply, **data)
File "/home/goback_main/anaconda3/envs/pl/lib/python3.8/site-packages/albumentations/core/transforms_interface.py", line 97, in call
return self.apply_with_params(params, **kwargs)
File "/home/goback_main/anaconda3/envs/pl/lib/python3.8/site-packages/albumentations/core/transforms_interface.py", line 112, in apply_with_params
res[key] = target_function(arg, **dict(params, **target_dependencies))
File "/home/goback_main/anaconda3/envs/pl/lib/python3.8/site-packages/albumentations/augmentations/transforms.py", line 2467, in apply_to_mask
return fn(mask, **params)
File "", line 52, in to_tensor
return x.transpose(2, 0, 1).astype('float32')
ValueError: axes don't match array

[Shawn94]

I`ve found out the solution to this problem:

1. You will need to expand dims to 1 of your mask images (512,640,1)
2. Normalize your mask image values. Normally in a single class, the values are either 255 or 0, so you should change it to 1 and 0, respectively.