Preannotations Not Visible in Object Identification Tasks

[nadinator]

Describe the bug I've set up my ML backend and connected it to an object detection project. The preannotations show up on the data view as the number 1, but when I actually click on the task

To Reproduce Steps to reproduce the behavior:

1. Create a project with the object detection template.
2. Set up a custom machine learning backend.
3. Connect the backend to the project and turn on all the settings in the Machine Learning settings page.
4. Navigate to the project's Data View (main page).
5. See that the Predictions column contains the number 1 (as opposed to 0, meaning that a prediction was made) for all tasks.
6. Click on any task.
7. See that there are no bounding boxes under the Regions tab.

Expected behavior I expected to see some bounding boxes representing the predictions of the backend that I could then accept and click Submit.

Screenshots

Environment:

• OS: Windows 10
• Label Studio Version 1.8.0
• label-studio-ml version 1.0.9

Additional context I'm using this model. I set everything up following the docs on the label studio website.

[turtle0x1]

I recently setup a backend ML and had similar issue, it was the way I was formatting the response annotations from the model. If your sure your model is making a prediction might be worth investigating!

[nadinator]

So it turns out it was making predictions. The prediction was simply empty, a.k.a no classes were found. I figured this out by looking at the results array in the task source. Might open a feature request to make this case clearer.

[nadinator]

While the results array is indeed empty, I'm actually no longer sure that this is the real reason. Gonna open this issue again and check out the response annotations as suggested.