Updating an existing object classifier

I am currently working on classifying cells in numerous slides reproducibly using the object classifier. Is there any way to give the classifier feedback on which cells he counted correctly and add the ones he missed to improve the automatization process?
I have read an article on github from 2 years ago (section " Interactively improve classification performance" in https://github.com/qupath/qupath/wiki/Classifying-objects) but am not sure if the described mechanism can be applied to the current object classifier somehow?!
Thanks in advance!

Hi @TMM-98,

Yes you can!
You might want to check the official docs for more info about this. Basically you can use the live update to check what the classifier got right/wrong and annotate further as required :slight_smile:

1 Like

Hi, this is what I already know - if I have understood your answer right. The thing I don’t know is if I can revisit the classifier after I have set it up initially and saved it.
So I have now saved a classifier which works okay (“Train object classifier” -> save). But is there a possibility to add future data points when I apply it to new images (“load object classifier”) in order to further enhance the detections?

If you have already understood my question right and answered it - I am sorry! But I have not found a way to reopen the training section of my saved classifier. If I really am too stupid would you mind giving me a hint to where I find this in the docs (they are my bible already…)

Thanks so much

The classifier itself is just the classifier, and I do not think it contains the training data. At least the last time I checked.

This is why it is good practice to keep separate images or even a separate project for training, so that you do not overwrite your training data. Then you can add to the training data and recreate the classifier at a later time.

In the old classifier, there was an option to save the training data (objects), which you could then load into a classifier along with any new data you annotated, but I do not think that is an option anymore.

As an aside, you can duplicate the project (so as to preserve the training data) just by copying the folder to a new location! Then overwrite your training data in the version you are working on all you want. If you go back and update the classifier based on new insights, you can copy the classifier from the Classifiers folder in the training project to the active project.

14 posts were split to a new topic: QuPath projects & version control