Followup for multi-annotation trainings

[daniel-wer]

Detailed Description

#8071 allows starting a job to train a model on multiple annotations, selected by id or uri. Follow-up wishes came up:

• When training a model from within wk, currently all bounding boxes of all annotations need to have the same size. This restricts which training data can be combined. Slight variations in size are usually no big deal and one even sometimes decides to "waste" some training data, because that's better than not using the data at all. Also, this becomes even more relevant with the next point. This will also require changes in the worker which should use the smallest bounding box of all to guide the training_sample_size computation (and related values). https://github.com/scalableminds/voxelytics/pull/3796 and Relax bounding box requirements for model training #8222
• It should be possible to select a resolution for each annotation which is the resolution that will be trained on. Many a time this is not the finest resolution (which is chosen currently). Only mags that exist both in the image data layer and the ground truth layer should be selectable. This is a frontend only issue.

Context

• Specific to long-running jobs (set jobsEnabled=true in application.conf)
• Specific to webknossos.org (set isDemoInstance=true in application.conf)

[daniel-wer]

@dieknolle3333 I updated the issue description to indicate that the first of the two TODOs was implemented already. I did not know this issue existed and was assigned to you. I hope you have not started working on that yet.

For the second TODO, @MatthisCl is implementing the worker side in https://github.com/scalableminds/voxelytics/pull/3833 and it would be great if you could make the mag selectable in the frontend as described in the issue :) To do so, please have a look at what I implemented in #8222. The warnings about the minimum bounding box size and that dimensions should be multiples will need to take the mag into account if it is different from 1 (as an example, if mag 2 is selected bounding boxes should be at least 20 vx [in mag 1] which corresponds to 10 vx in mag 2). Let me know if you have any questions about that.

[knollengewaechs]

@daniel-wer thank you for this update and the additional information. I didn't start working on this issue yet, but I am planning to take it on next :)

[knollengewaechs]

It should be possible to select a resolution for each annotation which is the resolution that will be trained on.

Does this only apply to trainings on multiple annotations?

[daniel-wer]

Does this only apply to trainings on multiple annotations?

It should also be possible when training on a single annotation (or from within an annotation).