how to determine sxy srgb etc. parameters in pydensecrf

[cliff007]

Hi all, I am not good at math, can you guys tell me how to set up a reasonable value of sxy srgb according to my own input image? Thanks!

• This adds the color-independent term, features are the locations only. d.addPairwiseGaussian(sxy=(3,3), compat=3, ……

• This adds the color-dependent term, i.e. features are (x,y,r,g,b). d.addPairwiseBilateral(sxy=(80,80), srgb=(13,13,13), rgbim=im, compat=10……

[cliff007]

I mean when to change the default setting of these parameters? I saw some setting is as below:

d.addPairwiseGaussian(sxy=20, compat=3) d.addPairwiseBilateral(sxy=30, srgb=20, rgbim=img, compat=10)

[lucasb-eyer]

Hi there,

there is no simple rule and it depends on your problem. Some people have developed an intuition over years of using this, but I haven't. Your best bet is to treat them as hyperparameters and to perform a search over them, with some way to test for the quality of the results.

Roughly speaking, they mean "is spatial proximity (sxy) or color similarity (srgb) more important? Is the pairwise-gaussian (compat=3) or the pairwise-bilateral (compat=10) more important?" and the answer to the questions depends on your data.

Here is one example where I did that, although without checking for the quality, just to give you a rough idea.

[huixiancheng]

hi,lucas,could u tell me the mean of the "lbp.progress" in your example? @lucasb-eyer is it mean the model Local Binary Pattern?

[MiguelAngeloMartinsRibeiro]

hi,lucas,could u tell me the mean of the "lbp.progress" in your example? @lucasb-eyer is it mean the model Local Binary Pattern?

Did you find what lbp.progress means?