Create a classification model using multiple images

Hi everyone,

I’m currently using Orbit to create a classifier for the cells of a lot of images. To create a well working model, I want to use a lot of different images to train the classifier. (approx 30). However, when I train a model, only the images that are opened, are taken into the training. Because I’m sharing my computer with some other people, a problem could occur when I close Orbit, because then all the shapes that I created for training the model will be lost. I know that I can save the model itself, but is there a way I can train one part of the images, close the computer, open Orbit again and continue with training the model? Or share the images including the training shapes?

Kind regards,
Johan

Dear @JohanPoort,
sure, you can train on a part of the images, then save the model.
Then later you can open the model and click on Model -> Restore Training Shapes
which will open all images used for the training so far and show the existing training shapes. You can then open further images and draw more training regions and click on train again (keeping all images open).
All used images must be available of course (either local or via image server).

Also for these kind of tasks the model explorer (right tabs) is useful. From time to time you can click on “add model” there to save a copy of your model to the clipboard which allows you to restore it later in case you’re not happy with the current version.

Hope that helps!

Regards,
Manuel

Dear Manuel,
I’ve tried to do so but it doesn’t seem to work. Opening the model works fine but when I click on ‘restore training shapes’ nothing happens. Have you got any idea what’s the problem?
Furthermore I was wondering what’s the best way to share annotations between computers. When i make an annotations, it saves on the database of the computer, but how can I share this with other computers?
Kind regards,
Johan Poort

Dear @JohanPoort,
are you using Orbit in stand-alone mode or connected to Omero?

For standalone the images must be available and it will only work on the computer the model has been trained.

Background:
Whenever Orbit sees a local image the first time a unique ID based on the MD5 hash of the image is assigned. Even if you move the images into another folder the ID remains.

For annotations: For Orbit standalone Orbit stores annotations (not training class shapes!) in the sqllite DB orbit.db in your user home.
This file can be copied to another computer (in the user home there) together with images. When you open images there, the annotations will also appear.

However, for a real collaborative workflow incl. training of images on several computers an image server like Omero is needed.
When Orbit runs in “Omero-mode” you can load class shapes and annotations across several machines.

Regards,
Manuel

Dear Manuel,

I’m using Orbit in stand-alone mode. I’ve created and saved a model yesterday using several images. When I reopened the model today and tried to restore the training shapes, nothing happened. I didn’t move or rename the images, although this shouldn’t matter like you said.

Thanks for the help on annotations!
Kind regards,

Johan

Dear @JohanPoort,
well, I think this is a bug when running Orbit in standalone mode. Can you please try to click on “Image -> Toggle Local/Remote Image Provider” and then on “Model -> Restore Training Shapes” again?

Regards,
Manuel