Why that number of anchor boxes?

Hello guys, I just want someone to explain like in general why in this example : https://www.mathworks.com/help/vision/ug/object-detection-using-yolov4-deep-learning.html they defined number of anchor boxes = 9...

and in this example https://www.mathworks.com/help/vision/ug/object-detection-using-faster-r-cnn-deep-learning.html they defined number of anchor boxes = 3...

Can someone tell me why did they use different number of anchor boxes?

Thanks for any response :)

Answer 1

[Birju Patel]

There isn't any rhyme or reason for these values. The examples need to be updated to provide more details on how to choose anchor boxes. I apologize for the confusion.

How you pick anchor boxes depends on your data.

As a first step, use this example to help you estimate the number and size of anchor boxes that would work well for your data:

https://www.mathworks.com/help/vision/ug/estimate-anchor-boxes-from-training-data.html

There will be a trade off between the number of anchor boxes and the complexity of the network. You should run experiments with different anchor boxes to zero in on the optimal values for your application.