0
$\begingroup$

Here's a tutorial about doing custom training of YOLO (Darknet): https://medium.com/@manivannan_data/how-to-train-yolov3-to-detect-custom-objects-ccbcafeb13d2

The tutorial guides how to set values in the .cfg files:

  • classes = Number of classes, OK
  • filters = (classes + 5) * 3

Why is it 'plus 5' then 'times 3'?

Some say it's (classes + coords + 1) * num, but I can't guess it out the meaning.

$\endgroup$
  • 1
    $\begingroup$ 3 boxes per feature map and for every box you predict class probabilities + width + height + x + y + object confidence. $\endgroup$ – Brale Feb 27 at 11:15
  • $\begingroup$ uhm, and what is that 'coords'? $\endgroup$ – datdinhquoc Feb 28 at 2:27
  • 1
    $\begingroup$ well its x, y, height and width $\endgroup$ – Brale Feb 28 at 8:57
0
$\begingroup$

As said by @brale in the comment below the question:

filters = (classes + 5) * 3
= (classes + width + height + x + y + confidence) * num
= (classes + 1+1+1+1+1) * num
= (classes + 5) * num

YOLOv3 dectects 3 boxes per grid cell, so it is:

filters = (classes + 5) * 3
| 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.