DGM lib. demo 'Train' question

Semantic Image Segmentation with Conditional Random Fields
wolfrun

DGM lib. demo 'Train' question

Postby wolfrun » Sun Apr 23, 2017, 20:41

Hello,
The DGM code really helps so lots of thing for my research. In fact, I try to implement CRF for semantic segmentation In your demo 'Train'.
Unfortunately I'm not an experienced graph model. It is hard to understand which model you do, only using your online documentation (Did you implement MRF by different training model??)
You also mention your papers in your project-X homepage but I failed to find same demo's pictures in your papers.
So could you refer for me which paper you did by Train demo? Or if you possible, could you give simple example about CRF?

User avatar
Creator
Posts: 128
Joined: Tue Dec 16, 2008, 21:52
Location: Hannover, Germany
Contact:

Re: DGM lib. demo 'Train' question

Postby Creator » Mon Apr 24, 2017, 21:38

Actually the same graph structure from Demo Train may be used for both CRF and MRF. The only difference between CRF and MRF is how you train your model, i.e. in the data for your potential functions (feature vectors). For MRF, the node feature vectors are calculated from a corresponding image pixel, and for CRF - the node feature vectors may be calculated from the whole image (like multi-scale features, variances, etc.)
Demo Train does not correspond to a specific publication, however papers
http://www.project-10.de/Kosov/files/ISPRS_2012.pdf
http://www.project-10.de/Kosov/files/PRRS_2012.pdf
may be a good starting point.


Return to “Direct Graphical Models”

Who is online

Users browsing this forum: No registered users and 1 guest