Philipp Lindenberger

Results 99 comments of Philipp Lindenberger

Hey, you can use [glue-factory](https://github.com/cvg/glue-factory) to train SuperPoint/SuperGlue or the free LightGlue on your dataset.

Hi again :) This seems to be a linking error between pycolmap and pixsfm. Did you check that also pycolmap is installed from source?

Hi :) Did you install COLMAP from source and run `make install`? `COLMAPConfig.cmake` should be located in `/usr/local/share/colmap/` or similar.

Hi, unfortunately, this is not possible right now. To make this work, one would need to build wheels for pixsfm, similar to how it is done in [pycolmap](https://github.com/colmap/pycolmap/tree/master/package). We are...

Hey @ericufatass 2000 images is quite a lot, and the demo matches pairs exhaustively, and you end up with over 1M pairs to match (thats completely unfeasible :D). Instead, you...

Hi @antithing, the current C++ part lacks an implementation of dense feature extraction and storage, thus running pixsfm entirely from C++ is not possible. However, if you assume to have...

Hi @sidk99, [match_dense](https://github.com/cvg/Hierarchical-Localization/blob/master/hloc/match_dense.py) replaces [extract_features](https://github.com/cvg/Hierarchical-Localization/blob/master/hloc/extract_features.py) and [match_features](https://github.com/cvg/Hierarchical-Localization/blob/master/hloc/match_features.py) from hloc, and results in the exact same output format, i.e. all the following steps (KA, mapping, BA) remain unchanged. And yes, no...

Hi @makmary, it is great to see hloc+pixsfm being used in practice! And sorry for the late reply. The objects in your scene have very little texture, so it is...

Sorry for the late reply. The error `/usr/bin/ld: /usr/local/share/colmap/../../lib/colmap/libpba.a(pba_generated_ProgramCU.cu.o): relocation R_X86_64_PC32 against symbol "_ZN3pba20jte_point_vec_kernelILi2ELi2EEEviiPf" can not be used when making a shared object; recompile with -fPIC` hints that pba in...

Hi @Xiaxia1997 Xiaxia1997, I have tested FeatureMetric KA on datasets with >1000 images and it worked, so this is seems to be a different problem. In KA, to build the...