Consult about DGM

Semantic Image Segmentation with Conditional Random Fields
Bryan

Consult about DGM

Postby Bryan » Sat Jun 03, 2017, 14:18

Hello, I am a graduate student and my major is road extraction from photo shot by small drone in the wild environment. I tried Kmeans clustering but the result didn't make me feel excited. One reason is that one photo of the clustering label to road is different from another photo. For example, the road label of the photo is number 1 and the road label of another photo is number 2 because the second one has different things like roof or parking which distract clustering. So I am ready to give up clustering. I don't know if the DGM could help me to extract road in the wild. Could you please tell me I could use which class or function in DGM to extract road? That could help me a lot.
Another question is that the DGM needs opencv3.2. Could I replace it with 2.4? Thank you!

User avatar
Creator
Posts: 157
Joined: Tue Dec 16, 2008, 20:52
Location: Hannover, Germany
Contact:

Re: Consult about DGM

Postby Creator » Sat Jun 17, 2017, 12:07

Dear Bryan,

I have just compiled the branch DGM opencv_2.4.13 with the OpenCV2.4.13. There were no mistakes with the Visual Studio 2015, however, this branch was derived few months ago, when the current version was 1.5.0, and thus it builds the libraries and binaries with prefix *150d.lib / *150d.dll

After you could build the DGM library, please use Demo Train demo as the starting point in road extraction. I suggest you to use classes CTrainNodeBayes or CTrainNodeCvRF for unary potentials and CTrainEdgePotts for pairwise potentials. Use the FEX module in order to extract features for your dataset. Section 4 of the paper describes a possible choice of the features.

I hope that helps you.


Return to “Direct Graphical Models”

Who is online

Users browsing this forum: No registered users and 17 guests