Closed zxrzju closed 6 months ago
Apologies about that; I realized I never pushed the splits! I've updated them in the data/OpenLane-V2/
folder.
The commands and checkpoints are still in the Training and Evaluating on Geographically Disjoint Set section. The data processing code should now work.
Hi, Thanks for sharing the great work! As is shown in the paper, training with the OpenLaneV2 splits may have overfitting. Will you provide the process_disjoint_split.py and the ckpts training with geo-disjoint splits for further evaluation?
Best regards