Handling empty detections while calculating PR curves

[shubham-scisar]

I ran inferencing on my image data,after getting results ,I received empty predictions on the data. Now when I am trying to generate PR curves, I have to calculate iou for the ground truth and predicted boxes. Now as the predicted boxes are empty and do not belong to any category. Can anyone suggest how to calculate iou for these empty detections ,as to how I can initialise these boxes with some value and class ? or should I just skip those images on which there is no detection and consider only those which have detections ? Kindly suggest.