1 Simplification 1
Observation
Simplification
Typical Setting
2 Simplification 2
Observation
Simplification
Two neurons with the same receptive field would not share parameters.
3 Benefit of Convolutional Layer
4 Another Story
Multiple Convolutional Layers
The channel of next filters is the last filters number of convolution
Although the filter is 33 on the second filter, but it consider 55 area on the ordinary picture.
5 Convolutional Layer
6 Pooling -Max Pooling
Convolutional Layers + Pooling
7 The Whole CNN
8 To Learn More
Data Augmentation: Flip, scale, crop, rotation
Spatial Transformer Layer to improve this problem.
