Open mfaramarzi opened 4 years ago
@homata @tosseto @yseki @KNSG @ksym27
I mentioned all the names but received no response.
Sorry for my late response. Japanese datasets are labeled with 8 classes. All you need to do is simply use 4 classes.
Sorry for my late response. Japanese datasets are labeled with 8 classes. All you need to do is simply use 4 classes.
@hiroyamaeda What are the crack types for remaining 4 labels?
Yes, Japan road dataset is made with 8 labels, but our task in this competition is to detect only four labels in that 8 labels. only D00, D10, D20, D40.
@mfaramarzi I think this issue can be closed.
all labels in the dataset are: Block crack D00 D50 D43 D40 D20 Repair D44 D10
@homata @tosseto @yseki @KNSG @ksym27 Many of annotations of "Japan datasets" have labels different than {D00: Longitudinal Crack, D10: Transverse Crack, D20: Aligator Crack, D40: Pothole}, as you can see below:
It seems those labels have not been updated from 2018 format (with 8 classes) to 2020 (with 4 classes.) Are we supposed to correct them one by one?