PyTorch-YOLOv3
PyTorch-YOLOv3 copied to clipboard
Why would get bbox , its one or two of xywh is inf, which cause the NMS a endless loop.
Should we add some restriction to the output of Yolo layer?
Merge overlapping bboxes by order of confidence
You can try to add 1e-16 to denominator. That works for me.
you should add code in utils/utils.py after large_overlap: if not any(large_overlap): detections = np.delete(detections, 0, 0) continue