You can now export the bounding boxes from your object detection dataset as cropped images usable with classification models. This update will enable easily prototyping two-pass models for use-cases like OCR and object tracking.
To try it out, simply enable the
Isolate Objects preprocessing step and generate a new version of your dataset.
This will generate one image (with one full-frame annotation) for each bounding box in the original dataset.
This allows you to export in the OpenAI Clip Classification format which converts your object detection dataset into a classification dataset. If you inspect the output, you'll see that your cropped bounding boxes are now organized into folders according to their class name.