Algorithm 1: The pseudocode of K-means

1: Given K cluster center points: $(W_i,H_i),i\in \{1,2,\dots ,k\}$,$W_i,H_i$ refer to the width and height of each anchor box.

2: Calculate the distance between each ground truth and each cluster center: $d(box,centroid)=1-\mathrm{IOU}(box,centroid)$. Since the position of the anchor box is not fixed, the center point of each ground truth is coincident with the clustering center.

3: Recalculate the cluster center for each cluster: $W^{\prime }{}_i=\frac{1}{N_i}{\displaystyle \sum w_i},H^{\prime }{}_i=\frac{1}{N_i}{\displaystyle \sum h_i}$

4: Repeat step 2 and step 3 until the clusters converge.