en
0.25
0.5
0.75
1.25
1.5
1.75
2
Latent Variable Models for Content-Based Image Retrieval and Structure Prediction
Published on Oct 09, 20127077 Views
In the first part of the talk I will present recent work on learning latent variable models for content-based image retrieval. To learn a function that predicts the relevance of a database image to a
Related categories
Chapter list
Latent Variable Models for Structured Prediction and Content-Based Retrival00:00
Scene Recognition (1)00:24
Scene Recognition (2)00:57
Why Hidden Variables? (1)01:19
Why Hidden Variables? (2)02:06
Outline03:03
Structure Prediction Problem 03:38
Structured Prediction03:39
Temporal Dependencies04:41
Spatial Dependencies05:49
Sequence Prediction Models (1)06:28
Sequence Prediction Models (2)07:30
Sequence Prediction Models (3)08:00
Representing distributions using WA08:53
Weighted Automata Representation09:12
Mapping from standard HMM to WA parametrization11:01
Spectral learning algorithm11:53
Why Spectral Learning?11:58
Hankel Matrix13:30
Duality between n-rank factorizations of Hankel and WAs15:45
Recovering Operators17:55
Spectral Method19:34
Discrete Homogeneous HMM20:58
Modeling paired sequences21:29
Experiments22:28
The features23:30
Results24:44
Mixture Model for Content-Based Image Retrieval (1)26:08
Mixture Model for Content-Based Image Retrieval (2)26:29
Complex Image Space26:54
Color important for outdoor images less important for indoor images27:30
Latent classes can model variability28:21
Global Ranking Model (1)28:32
Global Ranking Model (2)28:49
Global Ranking Model (3)30:42
Mixture Ranking Model33:19
Ranking model with latent variables (1)34:15
Ranking model with latent variables (2)36:30
Learning a Ranking Function37:01
Parameter estimation (1)37:05
Parameter estimation (2)37:52
Experiments38:11
SUN Dataset38:13
Ground-Truth Constraints40:20
Model Comparisons41:32
Results (1)43:16
Results (2)44:55
Latent Classes45:46
Summary & Future Directions47:00