The bench function returns negative FP for some samples, resulting in wrong F1 score calculation

[milad2073]

Hi,
When one of the predicted lanes can be matched to more than one ground truth the matched variable would be bigger than the number of predictions because of that, this line:
fp = len(pred) - matched
will calculate the FP negative.
Then the bench function returns negative FP for that image. This makes the calculated f1 wrong and then the final f1_score will be wrong.

I saw this issue in about 20 samples for my predictions and it resulted in a 0.09 percent error in the calculation of the F1 score.