Image Courtesy of Andrew Ng

My question is that how in YOLO, the networks does the midpoint of grid cell think ? I'm not completely sure I understand it. How can we know the midpoint of any object before actually knowing where the object is ?

Is it done via learning from the training data ? I can't completely seem to understand the training / learning process of the YOLO network. Any help would be appreciated.

here, YOLO is the paper You Only Look Once: Unified, Real-Time Object Detection by Joseph Redmon, Santosh Divvala, Ross Girshick, Ali Farhadi.

  • $\begingroup$ Please define YOLO. $\endgroup$
    – user234562
    Commented May 29, 2020 at 13:12
  • $\begingroup$ @user332577 Defined. $\endgroup$ Commented May 29, 2020 at 13:24

1 Answer 1


During the training, the ground truth bounding box is known, so one can easily compute its geometric centre and assign the object to the grid cell that contains this point. The loss function (eq. 3 in the original paper, also discussed in this question), for example, only considers a bounding box from thus identified grid cell but ignores the bounding boxes predicted by neighbouring grid cells, even if the object spills into them.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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