about the crf model

[huixiancheng]

about the crf model

[huixiancheng]

crf = PacCRF(21, num_steps=5, unary_weight=1.0)
crf.add_pairwise_kernel(kernel_size=5, dilation=1, blur=1, compat_type='4d', pairwise_weight=1.0)
edge_features = [paccrf.create_YXRGB(im, yx_scale=100.0, rgb_scale=30.0)]
output = crf(unary, edge_features)

dear ph.D, you use those code build a crf and then infer them.
however,i don't know the part of the build edge from two pairwise node since i'm noob could you tell me the part of the code? is it just like desecrf are all fullconnet?
or could i build a edge from two appointed points?

sry for in chinese:
尊敬的苏杭（？）博士：想问一下您的crf代码的边连接是参照densecrf进行全连接的吗？我能对指定两条点构建一条单独的edge吗？

[suhangpro]

The connection pattern is defined by the add_pairwise_kernel line. For example, kernel_size=5 means that each pixel is connected with all the other 5 * 5 -1 pixels within its neighborhood.

[huixiancheng]

thx,i will take a try