08. Pooling, VGG-16 Architecture
VGG-16 Architecture
Take a look at the layers after the initial convolutional layers in the VGG-16 architecture.
VGG-16 architecture
Pooling Layer
After a couple of convolutional layers (+ReLu's), in the VGG-16 network, you'll see a maxpooling layer.
- Pooling layers take in an image (usually a filtered image) and output a reduced version of that image
- Pooling layers reduce the dimensionality of an input
- Maxpooling layers look at areas in an input image (like the 4x4 pixel area pictured below) and choose to keep the maximum pixel value in that area, in a new, reduced-size area.
- Maxpooling is the most common type of pooling layer in CNN's, but there are also other types such as average pooling.
Maxpooling with a 2x2 area and stride of 2
Next, let's learn more about how these pooling layers work.