Stuck in levenberg_marquardt_strategy.cc loop for hours

[eyildiz-ugoe]

I have instructed a 3d reconstruction and it appears that it got stuck in LM strategy. My images had low number of features so I get why, but I must admit it's quite annoying that it spends hours just to fail afterwards. I see the below output only:

W0502 12:15:06.755776 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:06.829160 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:07.077844 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:07.153908 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:07.314445 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:07.474033 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:07.546308 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:07.792928 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:07.953227 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:08.028833 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:08.191509 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:08.351476 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:08.512042 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:08.585132 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:08.747380 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:08.907179 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:08.979694 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:09.314721 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:09.387681 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:09.548218 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:09.708550 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:09.784186 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.
W0502 12:15:10.031780 58541 levenberg_marquardt_strategy.cc:115] Linear solver failure. Failed to compute a step: CHOLMOD warning: Matrix not positive definite.

Is there a way to prevent this from happening? It kinda jeopardizes the whole reconstruction effort. Here is the command I use to launch the extraction:

run(colmap, f"feature_extractor --database_path '{work_dir}'/database.db --image_path '{work_dir}'/masked_images --ImageReader.single_camera 1 --ImageReader.camera_model PINHOLE --SiftExtraction.estimate_affine_shape=true --SiftExtraction.domain_size_pooling=true --SiftExtraction.use_gpu=true --SiftExtraction.gpu_index=" + str(gpu_index))

BTW it uses only 1 core of my machine while trying this. Ubuntu 22.04, GCC 10, as instructed.

Any idea what's going on?

[jytime]

It looks the correspondences (2D matches) are noisy and cannot lead to a reasonable solution. You can check libraries like hloc about how to use better detecting and matching methods instead of sift+NN.