GPS info during OpenSfM

[dakotabenjamin]

I was running the process with odm_data/seneca, which has GPS coords in the exif on all the photos (I double-checked), but I got this warning:

WARNING:__main__:Not all images have GPS info. Disabling matching_gps_neighbors.

I ran it on the master branch and did not run across this issue.

[edgarriba]

Not sure but I guess it's that flag. https://github.com/edgarriba/OpenDroneMap/blob/cmake/scripts/opensfm.py#L59

Anyways it's only a warning that's tells you what you said. Any consideration @paulinus ?

[dakotabenjamin]

I'm wondering if it has to do with https://github.com/OpenDroneMap/OpenDroneMap/commit/1cf79b64fffc0cd52198229b4099f6aad496cc5c

[dakotabenjamin]

What is happening in the declare_params() call? https://github.com/edgarriba/OpenDroneMap/blob/cmake/scripts/opensfm.py#L9

    def declare_params(self, params):
        params.declare("use_exif_size", "The application arguments.", False)
        params.declare("feature_process_size", "The application arguments.", False)
        params.declare("feature_min_frames", "The application arguments.", 0)
        params.declare("processes", "The application arguments.", 0)
        params.declare("matching_gps_neighbors", "The application arguments.", 0)

Is the last arg in params.declare supposed to be the default?

[edgarriba]

https://github.com/edgarriba/OpenDroneMap/blob/cmake/scripts/odm_app.py#L41

takes the value in 'matcher_k' which right I'm not sure if it's correct

[dakotabenjamin]

I'm currently updating those arguments to matcher_neighbors and matcher_distance. We'll see if it changes anything!

[edgarriba]

nice!

[dakotabenjamin]

well I'm not getting anywhere with the matcher-distance args, but I did find the source of the warning. Looking into it now: https://github.com/mapillary/OpenSfM/blob/odm-2/bin/match_features#L48

[dakotabenjamin]

Yeah I installed OpenSfM separately and ran bin/run_all, had the same warning. The config.yaml file has matching_neighbors, so all(map(has_gps_info, exifs)) must be returning false. My guess is the new directory design means that opensfm is not getting the exif info from the images. The image lists file uses absolute references to the images, I don't see how that could be the problem.

[paulinus]

There was a bug in that line checking for gps info. Just fixed that in OpenSfM odm-2 branch. Sorry for being slow here.

Btw, what will be the procedure for updating opensfm when using releases, instead of tracking a branch.

[dakotabenjamin]

That's a good question. How often do you expect OpenSfM to update such that it affects ODM?

[dakotabenjamin]

After that fix OpenSfM seems to be running at speeds like the master branch. I'll run some tests and PR an update.