Anchor boxes defined before boundary boxes

Hello everyone

I have very basic question. Why are anchor for defined before the boundary boxes. I understand that within boundary boxes can be two or more objects which can be defined by anchors. The question is related to Yolo algorithm for car detection.


The anchor boxes are input to the training here. They were precomputed by using unsupervised learning on the training data (k-means clustering on the existing bounding boxes in the training dataset). We are learning the bounding boxes based on the anchor boxes and the rest of the input data. The anchor boxes help the algorithm find the bounding boxes more accurately and efficiently.

Here’s a thread that discusses this in more detail.

1 Like

Thank you so much for very explanation. Now I have better understanding of the anchor boxes.