Abstract | ||
---|---|---|
We describe a probabilistic model for learning rich, dis- tributed representations of image transformations. The ba- sic model is defined as a gated conditional random field that is trained to predict transformations of its inputs using a factorial set of latent variables. Inference in the model con- sists in extracting the transformation, given a pair of im- ages, and can be performed exactly and efficiently. We show that, when trained on natural videos, the model develops domain specific motion features, in the form of fields of locally transformed edge filters. When trained on affine, or more general, transformations of still images, the model develops codes for these transformations, and can subsequently perform recognition tasks that are invari- ant under these transformations. It can also fantasize new transformations on previously unseen images. We describe several variations of the basic model and provide experi- mental results that demonstrate its applicability to a variety of tasks. |
Year | DOI | Venue |
---|---|---|
2007 | 10.1109/CVPR.2007.383036 | Minneapolis, MN |
Keywords | Field | DocType |
image coding,image motion analysis,image representation,statistical distributions,unsupervised learning,distributed representation learning,domain specific motion features,encodings,gated conditional random field,image transformations,locally transformed edge filters,probabilistic model,unsupervised learning | Affine transformation,Conditional random field,Computer vision,Pattern recognition,Computer science,Inference,Latent variable,Unsupervised learning,Probability distribution,Statistical model,Invariant (mathematics),Artificial intelligence | Conference |
Volume | Issue | ISSN |
2007 | 1 | 1063-6919 E-ISBN : 1-4244-1180-7 |
ISBN | Citations | PageRank |
1-4244-1180-7 | 85 | 6.77 |
References | Authors | |
20 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Roland Memisevic | 1 | 1116 | 65.87 |
geoffrey e hinton | 2 | 40435 | 4751.69 |