[Bug?] Class 0 small pixels count problem

[Ingerdev]

Mask annot\patient1\seq1\0039.png has 34 pixels labeled as class 0 and seems they don't mean anything
Mask annot\patient2\seq6\0023_l.png has 17421 pixels labeled as class 0 and they have a clean interpretation as "borders"

There are plenty of images where class 0 labeled pixels are less than 100.
Would you like to explain what it means and also name each of the 6 classes used in this dataset?

[mlaves]

Hi Ingerdev,
I had a look at the images you metioned and you are correct, that the class 0 pixels in seq1/0039.png are erroneous. We will reevaluate the images and correct them soon.
However, class 0 corresponds to void, meaning e.g. blurry borders between two classes or objects that cannot be categorized in the other classes. We used to weight class 0 with 0.0 in the loss function during training.
I additionally updated the description for the other class names.