search

cvg / sfm-disambiguation-colmap

Making Structure-from-Motion (COLMAP) more robust to symmetries and duplicated structures
280 stars 29 forks source link

with no file ".npy" #4

Open zr19980402 opened 2 years ago

zr19980402 commented 2 years ago

error

They say CANNOT load .npy file.

zr19980402 commented 2 years ago

I have found the wrong: when running the code, there is no track.

zr19980402 commented 2 years ago

After some attempts, the issue became a new. I cannot run the HLOC features. 微信截图_20220530215222 微信截图_20220530215239 微信截图_20220530215254 微信图片_20220530215312

lxxue commented 2 years ago

Can you post a minimal example to reproduce this error?

zr19980402 commented 2 years ago

First, I use the open dataset "books" as the input. other settings are shown in figure 1. 1

and I print the "output1" "output2" "output3" figure2-figure3 2 3

some details! 1.system: VMware Workstation Ubuntu20.04 2.Hloc version: Hierarchical-Localization-pairs_from_sequence ( This version only add the pairs from sequence)

  1. I have use the Recommended Hloc version as the same setting. the error is shown on the Last comment.
lxxue commented 2 years ago

From the error message in figure 2, it seems that the D2Net feature extractor is not installed correctly in hloc. I did a quick fix to make the codebase compatible with the new hloc interface. I pushed it to the branch new_hloc. Can you install the current version of hloc according to its install.md and try the codebase on this new branch?

zr19980402 commented 2 years ago

OK。I will try it at next day

zr19980402 commented 2 years ago

Thanks！ I have not seen this news, sorry.

zr19980402 commented 2 years ago

Make it! Thanks for your general answer！ Best wishes!

zr19980402 commented 2 years ago

And there may be a little misktake. In your /extract_match_features.py line 16-18~

assert feature_type in [ 'sift_default', 'sift_sparse', 'superpoint', 'r2d2', 'd2net', 'disk'

'sift_default', 'sift_strict', 'superpoint', 'r2d2', 'd2net', 'disk'

]
lxxue commented 2 years ago

Indeed it is a typo. Thanks for pointing it out!