Issues with "confirmed" images after deleting all annotations

[mary zotou]

Hello administrative team of CoralNet!

I am writing to ask your opinion regarding an issue I had with my source.

Initially, I had a source comprising >2000 images, of which ~1100 were confirmed. New labels were added, new classifiers were trained, and the known procedure was running smoothly. However, my research question changed, and I want to scrap the annotations and start from scratch (or almost from scratch). I want to keep the source, the images, the defined annotation areas of each image, and the metadata, but I want to delete all the annotations and start over. I selected the action "Delete Annotations Only" for the entire dataset, and now it seems like I still have >700 confirmed images. When I export the annotations of these "confirmed" images, the file is empty, and if I use the "Annotation Tool", there are no annotations in these images. Although in the "Jobs" Tab it says: Need 780 annotated images for next training, and currently have 709. But there are no annotations in these "confirmed" 709 images. How can I unconfirm these images? Is there a bug, or am I doing something wrong? 

Thank you in advance!

[Stephen Chan]

Hi Mary,

Regarding the number of confirmed images, the main page of your source should have a box on the right containing something like:

Image Status
Unclassified: [number]
Unconfirmed: [number]
Confirmed: [number]
Total images: [number]

This box's info should always be up to date.

On the other hand, what you see at the top of the Jobs page is not necessarily up to date. That shows the status at the time of the last source-check operation. But it's possible that the last source check happened before you deleted the annotations.

[mary zotou]

Hi Stephen,

The box has the same issue. It says 709 confirmed images.

I tried multiple times to delete all annotations, but nothing changed  in the main page box

[Stephen Chan]

Hi Mary,

I think I will have to look at this source to figure out what is happening. Could you give me the source ID or name? If it's a private source, you may send the ID or name to me privately ('Reply to author' in Google Groups).

[mary zotou]

It is a private source named: BioBoost

[Stephen Chan]

I tried deleting BioBoost's annotations myself, and it's exactly as you said. CoralNet says the operation completed, but 709 images are still Confirmed, yet those images do not actually have annotations. So the annotation-deletion part of the operation worked, but the image statuses failed to update, which seems to be a bug.

I'm aiming to fix the bug this week or early next week - thank you for reporting this.

[mary zotou]

Thank you for paying attention to my issue so promptly!

I would appreciate it if yoy try to fix it as soon as you can, since I am on a very tight deadline for the application I am working on.

[Stephen Chan]

I ended up fixing BioBoost's confirmed statuses manually. It should say 0 confirmed images now.

The underlying bug will take more time for me to figure out, since it's actually pretty complicated, requiring at least two processes to run with a specific timing. After the buggy behavior happened once, though, your next attempts at annotation deletion could not fix the incorrect image-statuses, because CoralNet no longer found any annotations to delete (and it will only update statuses if it finds annotations to delete).

Anyway, hopefully this situation doesn't happen very commonly and you can carry on with your work now.

[mary zotou]

Thank you so much for taking care of this issue!

Hopefully we won't need to start from scratch again...

[Diony Zervou]

Hello Stefen,

I am the admin in BioBoost. I tried to train a new classifier by uploading 20 image annotations. It keeps asking me to upload 780 image annotations. Is there a way to fix that?

Thank you!

Dionysia Zervou

[Stephen Chan]

Hi Dionysia,

Right, BioBoost still has trained classifiers based on the old annotations. The next training's confirmed-images requirement is based on how many confirmed images were used to train the last classifier.

To make the confirmed-images requirement go back down to 20, you'll have to delete BioBoost's previous trained classifiers. There isn't a one-click way to do this yet, but there are a couple of 'tricks' that work. The easiest way is to use the Add/Remove Labels page to add a label to BioBoost's labelset, save changes, then remove that same label from the labelset, and save changes again. CoralNet will then queue a job to delete the previous trained classifiers.

[Diony Zervou]

Hello again Stephen,

It worked! Thanks a lot, you will be mentioned in the "Acknowledgement" part of my thesis :)

Dionysia