You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
line in 125 : iou, bi = bbox_iou(pbox, tbox).max(0)
Why do you take the largest index value that overlaps an anchor with the GTs IOU? For example, when two objects are close together, if the two objects are IOU 0.98 and 0.93 respectively, but the categories of the two are different, 0.93 is correct. If the index is the largest, the correct answer is discarded.so i think it's not robust!
The text was updated successfully, but these errors were encountered:
@Peterisfar thanks for the comment. Yes you are right there is something incorrect in the test.py mAP computation. We left it this way because changing this causes the reported mAP to deviate further from the pycocotools mAP using python3 test.py --save-json.
line in 125 : iou, bi = bbox_iou(pbox, tbox).max(0)
Why do you take the largest index value that overlaps an anchor with the GTs IOU? For example, when two objects are close together, if the two objects are IOU 0.98 and 0.93 respectively, but the categories of the two are different, 0.93 is correct. If the index is the largest, the correct answer is discarded.so i think it's not robust!
The text was updated successfully, but these errors were encountered: