API/Import Feature in Frigate+ for Standardized Dataset Export/Import for Fine-Tuning (e.g., CVAT Integration)

[CarlosLimarino]

I'd like to propose an API import endpoint (or mechanism) that accepts a standardized dataset (e.g., COCO or Pascal VOC format) containing verified images with metadata (timestamps, camera, labels, etc.) for use in fine-tuning Frigate+ models.

This would allow users who prefer external annotation tools—like CVAT—to annotate their images outside Frigate+. With CVAT, I can run detection models to preclassify images and speed up manual annotation, thereby reducing load on Frigate+ server resources. An import feature would let users bring in these externally prepared datasets directly for model fine-tuning.

Currently, the Suggestions tab offers similar functionality, but an API-based import option would provide greater flexibility and support integration with any third-party tool.

I believe this feature would enhance the fine-tuning workflow and provide users more flexibility in managing their training datasets.

[blakeblackshear]

I haven't had many requests for this, so it's not likely to be a priority for a while.

I personally found CVAT and other annotation tools to be suboptimal when labeling thousands of images prior to launching Frigate+. It seems like they are designed to support all types of annotations, but aren't very efficient for any of them individually. My favorite tool was labelimg, which is now absorbed by Label studio.

I will likely invest time in improving the existing annotation tool rather than working on import/export features.

[CarlosLimarino]

Thanks for the response, Blake! I get that this isn’t a priority, but would you consider exposing an API endpoint that mirrors Frigate+’s existing annotation functionality? This way, users could programmatically upload annotated data without requiring a new format or additional complexity.

Would love to hear your thoughts—thanks again!