[Help Wanted](Per Class IU Calculation) Problem in utils>metric_util.py

[Guru-Deep-Singh]

Dear Authors,

In the function fast_hist_crop, I see that you are removing the 0th row and 0th column before calculating the IOU. However, this will remove the points predicted 0 for ground truth other than 0 and also all the points which have ground truth 0 and are predicted something else.

I think the code should be changed to :
def fast_hist_crop(output, target, unique_label):
hist = fast_hist(output.flatten(), target.flatten(), np.max(unique_label) + 2)
hist = hist[:-1, :]
hist = hist[:, :-1]
return hist

This will remove the last column and last row. This will remove the row and column of the class which is not in the unique value of classes.

[Guru-Deep-Singh]

Hello Authors! Can you please look into my query.

Thank you

[xinge008]

@Guru-Deep-Singh In my opinion, the predicted 0 is the ignore. So we just follow existing methods to remove these predictions. You can also see it in many previous work, like PolarSeg, SalsaNext, etc.