Error when training with my own data about hfnet HOT 3 CLOSED

Hi,

It seems unlikely that you are trying to match against 182649 descriptors, so ind should probably not have such high values. Could you check the size of dist? If that size is <182649, looks like there could be some overflow issues in topk.

By the way, we found that the localdb_hfnet.pkl is much larger than localdb_superpoint.pkl, is that right?

No, that shouldn't be the case, the number of keypoints should be identical, and the descriptors have the same dimensionality. This issue could be related to the error that you describe.

from hfnet.

Thank you for your advice, the problem has been solved.

from hfnet.

Thank you for your advice, the problem has been solved.

I have the same issue, what was causing the issue in your case ?
I am using a split of RobotCar in my case (sequences 004 040 and 044).

from hfnet.