Experiment 15
Learning Hierarchical Active Basis Model from Roughly Aligned Images (being refined and expanded by Zhangzhang Si)
The hierarchical
active basis model shares the same form as the original active basis model. It is a simplest recursive two-layer AND-OR graph.
The hierarchical
active basis model consists of several sub-templates, each sub-template is represented by an
active basis model, and each sub-template is allowed to shift its overall location, orientation and
scale. It is related to Experiment 5.d
and Experiment 7.
(1)
data, code, and readme for cats example
The learned plates and their log-likelihood colors.
Deformed hierarchical active basis matched to the training examples.
(2)
data, code, and readme for egrets example
The learned plates and their log-likelihood colors.
Deformed hierarchical active basis matched to the training examples.
(3)
data, code, and readme for bikes example
The learned plates and their log-likelihood colors.
Deformed hierarchical active basis matched to the training examples.
Back to active basis homepage