mihaidusmanu/d2-net

About training another dataset

Jo-won opened this issue · 2 comments

Thank you for good paper and code.
I will use this model in another datasets.
About that, i wonder whether i should re-train about another datasets or just use checkpoint about mega-depth in this case.
So could you tell me whether to re-train or not, when you evaluate in InLoc and Aachen Day-Night dataset?

For all our experiments we trained only on MegaDepth and then used those weights to evaluate on HPatches Sequences / InLoc / Aachen / etc. No retraing was done.

Thank you for your answer!