Question: Integrating a Fine-tuned DETR Model for Object Detection

Hi HumanSignal team👋
First and foremost, thank you for all your hard work on Label Studio and its ML backend. It's a great tool for the data science community, and I'm very grateful for the effort you've put into open-sourcing it. 😻🌟

I have a question regarding the integration of a custom model:

I've fine-tuned a DETR-based model for a specific object detection use case and would like to integrate it into the Label Studio ML backend. I've successfully run the provided YOLO-based example and have also attempted to adapt the existing Hugging Face BERT example, though understandably, neither was directly applicable to my object detection model.

Could you provide some guidance or point me towards the recommended approach for integrating a custom DETR model for object detection within the Label Studio ML backend? Are there any specific patterns or examples you would suggest reviewing beyond the above mentioned ones?

Any insights 💡 or recommendations would be greatly appreciated!

Thanks a lot,
Chris