raise ValueError('Non-used anchor #{} found'.format(aind))

[wenglihong]

Hello, I converted lyft open dataset to Kitti format and tried to train on the converted dataset, but I got error "raise ValueError('Non-used anchor #{} found'.format(aind))" on rpn_util.py, after some debugging, I found the problem is that some anchors were never chosen as max prob target anchor, so I want to know how to adjust hyper parameters to fix the problem, I think it's quite tricky to choose some parameters that can make each anchor has probability to be the max prob target anchor, and how these anchors will affect the performance. Hope to get some ideas from you or guys here

[cnexah]

I think you can remove some anchors, such as removing some kind of aspect ratio

[Sondosmohamed1]

Hello, I converted lyft open dataset to Kitti format and tried to train on the converted dataset, but I got error "raise ValueError('Non-used anchor #{} found'.format(aind))" on rpn_util.py, after some debugging, I found the problem is that some anchors were never chosen as max prob target anchor, so I want to know how to adjust hyper parameters to fix the problem, I think it's quite tricky to choose some parameters that can make each anchor has probability to be the max prob target anchor, and how these anchors will affect the performance. Hope to get some ideas from you or guys here

hello i have the same problem did you solve your problem? i generated the kitti dataset from Carla and I got this exception