Convolutional Networks
Filters
CNN’s Topology
Feature maps
Feature extraction layer
Convolution layer
C
Shift and distortion invariance or
Subsampling layer
61
S
Feature extraction
 If a neuron in the feature map fires, this corresponds to
a match with the template.
62
Subsampling layer
 the subsampling layers reduce the spatial resolution of
each feature map
 By reducing the spatial resolution of the feature map,
a certain degree of shift and distortion invariance is
achieved.
63
Subsampling layer
64
Deep Convolutional NN for Image Recognition
CNN: local connections with weight sharing;
pooling for translation invariance
2012-2013
Fully connected
Fully connected
earlier
SVM
Pooling
Histogram Oriented Grads
Image
Fully connected
Convolution/pooling
Convolution/pooling
Convolution/pooling
Convolution/pooling
Convolution/pooling
Raw Image pixels
Learning a Compositional Hierarchy of Object Structure
Parts model
The architecture
Learned parts