Minimum detection box size for Object Detection


I’m wondering if there is a proven way to find the minimum detection box size for a given object detection model. I’m hoping to figure this out so that I can build a dataset with samples that reach this limit but do not pass it.

If there is no such thing as this minimum detection box for an object detection model, I’m guessing that it is probably “created” by including small objects in the training dataset.

Specifically, I’m wondering about the SpaghettiNet large model.