Closed Willyzw closed 2 years ago
Thank you for pointing this out. This is indeed a bug and could cause an illegal memory access, but seems to have negligible impact on the accuracy of our system. And it only affects inference, not training or evaluation.
Anyways I am going to correct this issue. Our results may improve slightly after fixing the bug, and we are running some tests to check.
Great! Since you have correct the index in the last commit, I am closing this issue.
Thanks for releasing this excellent work! One small thing drew my attention and unsure if this is on purpose, which is the
xi[45]
at this line. https://github.com/princeton-vl/DROID-SLAM/blob/9843482ec70a84d28defe3def4ed84b3a77ce914/src/droid_kernels.cu#L154