0.25
0.5
0.75
1.25
1.5
1.75
2
Improved Initialisation and Gaussian Mixture Pairwise Terms for Dense Random Fields with Mean-field Inference
Published on Oct 09, 20125059 Views
Recently, Krahenbuhl and Koltun proposed an efficient inference method for densely connected pairwise random fields using the mean-field approximation for a Conditional Random Field (CRF). However,
Related categories
Chapter list
Improved Initialisation and Gaussian Mixture Pairwise Terms for Dense Random Fields with Mean-field Inference00:00
Labelling Problem00:03
Problem Formulation (1)00:12
Problem Formulation (2)00:22
Problem Formulation (3)00:49
Inference in Dense CRF (1)01:07
Inference in Dense CRF (2)01:29
Inference in Dense CRF (3)02:04
Naïve mean field02:31
Efficient inference in dense CRF02:45
Marginal update03:05
Q distribution (1)04:01
Q distribution (2)04:11
Q distribution (3)04:12
Q distribution (4)04:13
Q distribution (5)04:20
Two issues associated with the method04:28
Our Contributions04:59
Sensitivity to initialisation05:22
SIFT-flow based correspondence (1)06:07
SIFT-flow based correspondence (2)06:27
SIFT-flow based correspondence (3)06:40
Label transfer06:47
SIFT-flow based initialisation (1)07:20
SIFT-flow based initialisation (2)07:53
Gaussian pairwise weights (1)08:18
Gaussian pairwise weights (2)09:04
Our model09:34
Learning mixture model (1)10:19
Learning mixture model (2)10:29
Learning mixture model (3)10:43
Our model (2)10:48
Learning mixture model (4)11:17
Inference with mixture model11:49
Experiments on Camvid (1)12:25
Experiments on Camvid (2)12:43
Experiments on Camvid (3)12:45
Experiments on Camvid (4)12:55
Experiments on Camvid (5)12:56
Experiments on Camvid (6)13:11
Experiments on Camvid (7)13:51
Experiments on PascalVOC-10 (1)14:11
Experiments on PascalVOC-10 (2)14:45
Experiments on PascalVOC-10 (3)15:57
Conclusion16:24
Thank you17:19