Issue with Pre-Labeling Using LabelBox Model in Batch Creation on Labelbox

Hello,

I’m Javier Pérez, working on the SolarPanelsAI project. I’ve developed a custom model within Labelbox for pre-labeling, but I’m facing an issue. When creating new batches, the option to use my model for pre-labeling is not appearing, even though the documentation suggests this should be possible.

Has anyone else encountered this, and could you provide guidance on enabling this feature? I’m looking to submit my model’s predictions as pre-labels in the batch creation process but am unable to find the option to do so.

Any advice or suggestions would be greatly appreciated!

Best,
Javier

Hello @javipg99 ,

Do you have predictions (not ground truth) in your model run(s)?
Are those predictions matching the schema ontology set for those predictions and the destination project ?

If those 2 are correct, please send me a project id, I can have a look.

Many thanks,
PT

Hello PT,

Thank you for your fast response.

Yes, the predictions from the model runs are indeed not ground truth and they match the schema ontology set for the predictions and the destination project. The issue arises when I try to create a new batch with the second import, which contains the remaining 121 images that need labeling. The first 72 was made manually are already utilized for the model training within Labelbox. However, when importing this new batch into the project, the option for pre-labeling with the model’s predictions does not present itself.

The project ID in question is clp3xmypy0fhs072d4zwd3771. I’m trying to apply the pre-labels to facilitate the labeling of these remaining images and enhance the model’s training further. Any assistance or insights you could provide would be immensely helpful.

Many thanks,
Javier Pérez

Hey @javipg99,

So just for clarification are you trying to use a custom model to pre label images within the UI (labelbox platform itself)

Thank you!

Hey there,

Yes, that’s correct. I am attempting to use a custom model that was created and trained within Labelbox to pre-label images directly from the Labelbox UI. This model has been trained on images and annotations that were manually annotated within Labelbox.

Thank you for looking into this for me!

Best,
Javier Pérez

Hey @javipg99,
Unfortunately, foundry is not released yet, which is what you are referencing. The open beta is planned to release tomorrow, but it will only include models hosted by Labelbox for pre-labeling, which would be models like Amazon image recognition, GPT3, and Yolo. The ability to use custom models to pre-label data rows through our UI, in your case, is slated to be release early next year. I would keep an eye out on our blog for any further news.

You can import annotations from a custom model through our SDK. You would just have to run each data row through your model and then import the result. Here is an article on images but most media types are supported. Import image annotations

Best,
Gabe

Hey @javipg99, Foundry is now released! Feel free to check out this community post and check out the guides on how you can use Foundry for your projects: Introducing Foundry