Hello
I am confusing about the code, I readed the article "https://arxiv.org/pdf/1612.08242.pdf" "Page 3" and I was surprised that the author mentioned the k-means algorithms for generating the bounding box, and it's was mentionned that it is better than generate bouding box by hand
My question is where is situated the part that implement the k-mean
Thanks
K-mean is used to calculate anchors, which is already provided in .cfg files, so if you had your .cfg files (which darkflow assumes); then there is no need to re-compute the anchors and thus, no need for use of k-mean.
Hi @thtrieu Could you explain what the anchors = 0.57273, 0.677385 ... in .cfg files mean for bonding box? For example, what is anchors = 1, 1 corresponding to the size in the real image for YOLOv2 416X416.
Most helpful comment
K-mean is used to calculate anchors, which is already provided in
.cfgfiles, so if you had your.cfgfiles (whichdarkflowassumes); then there is no need to re-compute the anchors and thus, no need for use of k-mean.