search

cvg / pixel-perfect-sfm

Pixel-Perfect Structure-from-Motion with Featuremetric Refinement (ICCV 2021, Best Student Paper Award)
Apache License 2.0
1.29k stars 132 forks source link

Running dense matching from scratch in demo #97

Open sidk99 opened 1 year ago

sidk99 commented 1 year ago

Thanks for the great work, a few questions: 1) Current setup for dense matching using loftr is done for the eth3d dataset and seems to be designed to run on an preexisting colmap model. Can the dense_matching be used from scratch to create pairs without any other feauture extraction/pairing? 2) How can the match_dense be used in the sample demo provided?

Phil26AT commented 1 year ago

Hi @sidk99, match_dense replaces extract_features and match_features from hloc, and results in the exact same output format, i.e. all the following steps (KA, mapping, BA) remain unchanged.

And yes, no other feature extraction / pairing is required.