16.03.2021 Views

Advanced Deep Learning with Keras

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Deep Neural Networks

Generally speaking, we can observe that the layer l output feature maps are directly

affected by the previous feature maps only. Meanwhile, for ResNet:

x

( x )

l-1 = H

(Equation 2.2.3)

l-2

( ( -1 ) -2 )

x = ReLU F x + x

(Equation 2.2.4)

l l l

F

( )

Figure 2.2.3: A detailed layer operations for a plain CNN block and a Residual block

x

l−1

is made of Conv2D-BN, which is also known as the residual mapping.

The + sign is tensor element-wise addition between the shortcut connection and

the output of F ( x

l−1

) . The shortcut connection doesn't add extra parameters nor

extra computational complexity.

[ 52 ]

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!