caffe

caffe copied to clipboard

Hi Wei,

Thanks for sharing the code! Quick question about the steps parameter. It seems to me the step should be a power of 2. In your code, steps is set as follows steps = [8, 16, 32, 64, 100, 300] What's the logic behind 100 and 300? If I want to add a couple of more conv layers, how should I choose the steps?

Because conv8_2 and conv9_2 has stride of 1 instead of 2.

what is steps used for? what does it have to do with the stride?

steps are used to generate prior box.

why not use feature_size = [38,19,10,5,3,1] derectly? just like said in the paper, like

for x in range(feature_size[i]):
        for y in range(feature_size[i]):
                cx = (x+offset)/feature_size[i]
                cy = (y+offset)/feature_size[i]

but use step to get the approximate, like 300/38=8, 300/19=16, ... , 300/1 =300, to get steps to generate prior box?