Hi, thanks for this excellent work!
I try to train the model using the config: R50_FPN_daytime_cycle_conf.yaml on BDD100k.
However, the performance is only 18.126 AP which is the same as my retrained baseline (18.108 AP).
There is a 1.0 AP gap with the reported performance in the paper.
Hi, thanks for this excellent work! I try to train the model using the config: R50_FPN_daytime_cycle_conf.yaml on BDD100k. However, the performance is only 18.126 AP which is the same as my retrained baseline (18.108 AP). There is a 1.0 AP gap with the reported performance in the paper.