Yes, we have. But we find that since we follow the dataset split of superglue, we put the scene of Megadepth-1500 into the training set, so we do the Megadepth evaluation on our private setting(the scenes which are not in the training set of superglue, PATS, and LoFTR)
Yes, we have. But we find that since we follow the dataset split of superglue, we put the scene of Megadepth-1500 into the training set, so we do the Megadepth evaluation on our private setting(the scenes which are not in the training set of superglue, PATS, and LoFTR)