mihaidusmanu/d2-net

InLoc dataset

Closed this issue · 4 comments

bareblackfoot commented

How can I evaluate your code with InLoc dataset? (visual localization)

mihaidusmanu commented

Details regarding the InLoc evaluation are available in the inloc directory at: https://github.com/mihaidusmanu/d2-net/tree/master/inloc. In case you already tried that and ran into issues, please let me know.

bareblackfoot commented

Details regarding the InLoc evaluation are available in the inloc directory at: https://github.com/mihaidusmanu/d2-net/tree/master/inloc. In case you already tried that and ran into issues, please let me know.

Thanks a lot! I followed the instruction.
But I am stuck in converting results to txt format file (imglist2text.m)
Would you let me know what is the right input of that function?

The code needs a 'queryname' and 'P', but I think the output of your algorithm provides only keypoints, descriptors, and scores.

mihaidusmanu commented

As mentioned in the readme, in order to get the poses, you will need to run custom_demo which will use the InLoc evaluation pipeline with D2-Net features (instead of SIFT / Dense). Upon completion of this step, you will find a .mat file with poses that can be fed toimglist2text.

  • 👍1
bareblackfoot commented

Oh, I got it. Thank you a lot again :)