1
$\begingroup$

I know how IOU works during detection. However, while preparing targets from ground truth for training, how is the IOU between a given object and all anchor boxes calculated?

Is the ground truth bounding box aligned with an anchor box such that they share the same center? (width/2, height/2)

I think this is the case but I want to hear from someone who has better knowledge of how training data is prepared for training in YOLO.

$\endgroup$
  • 1
    $\begingroup$ Can you perhaps provide a bit more context about the system you are talking about? $\endgroup$ – Oliver Mason May 22 '18 at 14:03
  • 1
    $\begingroup$ @OliverMason: Here's what I am talking about: pjreddie.com/darknet/yolo In order to train, I need to prepare the training data. One of the fields is 'objectiveness factor' which is defined as IOU between the actual bounding box and best-fitting anchor box. My question is: how are these two boxes aligned? $\endgroup$ – nivter May 22 '18 at 14:19
1
$\begingroup$

My assumption was correct: the ground truth bounding box is aligned with an anchor box such that they share the same center

In other words, only the widths and heights are used to calculate the ground truth IOU.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.