RockStarCoders / alienMarkovNetworks

Using MRFs and CRFs for computer vision problems.
21 stars 9 forks source link

brainstorm and read up on possibilities for edge potentials #25

Closed jsherrah closed 11 years ago

jsherrah commented 11 years ago

are we using all the info we could? make them more data driven?

amb-enthusiast commented 11 years ago

I'll read up on the co-occurrence approaches used by Ladicky 2013 & Souiai 2013.

jsherrah commented 11 years ago

And I Yao and Rubinstein (see #21 )

amb-enthusiast commented 11 years ago

Updated (https://github.com/RockStarCoders/alienMarkovNetworks/issues/21) to include description of Ladicky et al Co-occurrence CRFs paper.

Seems we can use the co-occurrence cost C(L) and apply existing Boykov approximate graph cuts algos for inference on the resultant CRF. I think this could be a good next step when we have the baseline model in place.