joydeepmedhi / anchor-boxes-with-kmeans Goto Github PK

.

Hi Joydeepmedhi,

Thanks for these codes. They are very helpful.

Can i ask why the formula for Area Scale is
data['b_area_scale'] = (data['w']*data['h']/(base_box)).apply(np.sqrt)
instead of
data['b_area_scale'] = (data['b_w']*data['b_h']/(base_box)).apply(np.sqrt)?

and why the for formula for TF Scale is
data['tf_scale'] = data['b_h'](data['b_ar']).apply(np.sqrt)/base_anchor
instead of
data['tf_scale'] = data['b_h'](data['b_ar'])/base_anchor?
(why is square root needed for TF Scale?)

I am sorry to bother you. I would like to ask how these clustering results can be used in Faster-RCNN. I see a lot of output in your .ipynb. Can I replace the corresponding parameters in Faster-RCNN config file with the clustering results such as Aspect Ratios and Scales? I have just come into contact with this research direction. I look forward to your reply, thank you!

Why do you set such values in input_array=[64,96,128,196,212,256,512]? how do you get that, like 212?