bob.db.multipie issues
https://gitlab.idiap.ch/bob/bob.db.multipie/-/issues
2017-11-16T06:23:08Z
https://gitlab.idiap.ch/bob/bob.db.multipie/-/issues/4
Training set of protocol 'P' includes cameras '19_1' and '08_1', while test s...
2017-11-16T06:23:08Z
André Anjos
Training set of protocol 'P' includes cameras '19_1' and '08_1', while test set does not
*Created by: siebenkopf*
When I read the create script fro the protocols:
https://github.com/bioidiap/bob.db.multipie/blob/master/bob/db/multipie/create.py#L306
I see that the training set for all protocols are composed of all cameras...
*Created by: siebenkopf*
When I read the create script fro the protocols:
https://github.com/bioidiap/bob.db.multipie/blob/master/bob/db/multipie/create.py#L306
I see that the training set for all protocols are composed of all cameras (``cams_all``). The probe sets of the protocols consist only of the provided ``cams``, which are ``cams_all_P`` for protocol ``P``:
https://github.com/bioidiap/bob.db.multipie/blob/master/bob/db/multipie/create.py#L317
I think I remember that we have removed the cameras ``19_1`` and ``08_1`` (the difference between ``cams_all`` and ``cams_all_P``) from the ``P`` protocol, as these two cameras were used in awkward angles (and the annotations might have been mixed up). Now, the question is, why do we have these cameras in the training sets of all ``P*`` protocols, including the ``P`` protocol?
Should we (finally) remove the two cameras also from the training set of the ``P`` protocol (and maybe from the other ``P*`` protocols as well)?
A related question would be, if it would make sense to use only the desired cameras for the training sets of the ``P*`` protocols, i.e., the actual camera for the given pose, and the ``05_1`` camera, which is used for enrollment?