How to know a certain detection is under reasonable or partial or heavy occlusion from the converted xml file?
jundeli opened this issue · comments
After conversion, i checked the occlusion label, it shows only 0 and 1. How do we distinguish partial occlusion (occlusion range less than 35%) and heavy occlusion (above 35%) from the converted file? If I only want to consider reasonable detections, should I only pick bbox with occlusion of 0 from the xml file? And how about partial and heavy occluded cases?
@jundeli I think you could calculate the IOU values between each box in the same image and then select a proper threshold to distinguish the partially occluded objects and heavy ones. But the most important thing is to review the target category images directly. It could help to optimize the rule to select the best images and bounding boxes.