DetectionDataset accumulator

Search before asking

• I have searched the Supervision issues and found no similar feature requests.

Description

This feature is interested for semi-automatic annotations. User can use this feature to use their favourite pretrained model to generate detections and then with sv.DetectionDataset can magically store into favourite format.

import supervision as sv

ds = sv.DetectionDataset.empty(...)

image =  cv2.imread(`image-path`)
inference_output = model(img)  # can be anything i.e. SAM, Dino, transformers, yolo, etc
detections = sv.Detection.from_model(inference_output)

image_path = "1.jpg"
ds.add_detection(image_path=image_path, image=image, detections=detections)

ds.as_coco(output_image_directory, annotation_path)

Use case

• Semi-automatic annotations
• Annotations corrections

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

Start by locating DetectionDataset, Detection.from_model, and the existing dataset export APIs mentioned in the example; inspect how images, detections, and annotation formats are currently represented. The feature is complete when a dataset can accumulate model-generated detections and export them in the supported formats for semi-automatic annotation workflows.