Support for COCO Annotation Format #230
Description
Is your feature request related to a problem? Please describe.
Annotate-lab currently supports YOLO-format annotations, which works well for many workflows. However, many datasets, tools, and ML pipelines rely heavily on the COCO annotation format (JSON-based). The lack of COCO support makes it challenging to integrate annotate-lab into existing ecosystems or export annotations for tools that require COCO format.
Describe the solution you'd like
I would like annotate-lab to support both importing and exporting annotations in the COCO dataset format. Specifically:
- Ability to upload COCO JSON annotation files and map them to the corresponding images.
- Ability to export completed annotations as a valid COCO JSON file.
- Support for bounding boxes, segmentation polygons, and masks (if applicable).
- Proper UI visualization/editing for COCO-style annotations.
Describe alternatives you've considered
- Using external converters or custom scripts to translate between YOLO and COCO formats, which adds unnecessary complexity and risk of conversion errors.
- Continuing to use YOLO format exclusively, but this limits compatibility with training frameworks like Detectron2 and many benchmark datasets.
- Building a custom conversion pipeline, which is not scalable for larger or collaborative projects.
Additional context
Supporting COCO format would significantly enhance annotate-lab’s interoperability with modern ML tooling, including Detectron2, SAM/SAM2 pipelines, and widely available COCO-based datasets.
This feature would also help broaden annotate-lab's adoption among researchers, practitioners, and annotation teams who rely on COCO as the standard annotation structure.