Member

Works like a charm on the datasets given in https://github.com/json87/SphereSfM.

What can be improved:

• make render_perspective_images faster (currently 1-2s per pano) by moving the logic to C++ (though cv2.remap is already multi-threaded).
  • We could create a new CLI steps that outputs a rig_config.json to be consumed after feature extraction by the rig_configurator (unfortunately this can't be done in a single step).
• add an option to make feature matching ignore pairs between images of the same frame (since we know here that they are not covisible)
• geometric verification with rig constraints (I have a quick prototype based on PoseLib, for another PR)

Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Contributor

Great, thanks very much. Shall we also add reference this example in docs/rigs.rst?

Member Author

I added a short doc. I will try to record a small dataset (with GPS pose priors) and add it to the example.

Contributor

Just gave it a try. This works smoothly in my end. Thanks for the great example script!