Closed ppbangKGT closed 1 year ago
@ppbangKGT Thanks. I think your question has two parts. Refer to: https://github.com/OpenDriveLab/OpenLane/blob/main/anno_criterion/Lane/README.md
track_id
. Since we generated the whole landline first and then crop by perception field for each sample.attribute
in annotations. Note that we only have attribute numbers for LL/L/R/RR four lanelines in the ego direction. We excluded opposite-direction lanelines since we thought they were not used for decisions in most cases. The 1/2/3/4 numbers are just easily assigned. No particular reason for this and you can just change them to 1/3/2/4, etc. This part is probably not that consistent with industrial methods.
Thanks for your nice work! When I think about the lane tracking task, I wonder how you annotate the tracking id. It's obvious that the id should be unique in one segment. But how you decide the labeled number? I've seen some datasets that number the same direction lane lines as 0, 1, 2, 3... in order, and number the opposite direction lane lines as -1, -2... in order. Some datasets also number the left lane lines as 1, 3, 5... with the current lane as the center, and number the right lane lines as 2, 4, 6... So what is your numbering method? Thanks for your time.