Towards removing custom glomap image class

[ahojnnes]

First step in removing the custom class. Next steps will then remove the features/features_undist fields as well as refactor the current "misuse" of "IsRegistered/HasPose" for flagging images in the largest connected component.

[B1ueber2y]

LGTM. From reading the code there seem to be no behavior changes.

We can largely benefit from a benchmarking similar to the incremental mapper of colmap to ensure stability. Or alternatively, a faster option is that we can have some reference reconstructions on some scenes from the original glomap and check reconstruction statistics to ensure equivalence.

[ahojnnes]

Agreed, see other PR.

[B1ueber2y]

Next steps will then remove the features/features_undist fields as well as refactor the current "misuse" of "IsRegistered/HasPose" for flagging images in the largest connected component.

Does it make sense to refactor the misuse in this PR? After merging it would be difficult to locate as IsRegistered / HasPose will be mixed up in the codebase.

[ahojnnes]

Next steps will then remove the features/features_undist fields as well as refactor the current "misuse" of "IsRegistered/HasPose" for flagging images in the largest connected component.

Does it make sense to refactor the misuse in this PR? After merging it would be difficult to locate as IsRegistered / HasPose will be mixed up in the codebase.

I am not sure this PR makes it more difficult. You can just grep all usage of HasPose() in the src/glomap subfolder to refactor.