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

Transition Layer

(2)

Residual Block

(3)

16 × 16 { 1× 1 Conv2 D, strides = 2}

8 × 8

8 × 8 64 ⎧3× 3 Conv2 Dstrides , = 2if 1st Conv2D⎫ ⎨

⎬ × n

⎩3×

3 Conv2D

Average Pooling 1 × 1 8×

8 AveragePooling2D

Table 2.2.1: ResNet network architecture configuration

Figure 2.2.4: The model architecture of ResNet for the CIFAR10 dataset classification

# Layers n % Accuracy on CIFAR10

(Original paper)

ResNet20 3 91.25 92.16

ResNet32 5 92.49 92.46

ResNet44 7 92.83 92.50

ResNet56 9 93.03 92.71

ResNet110 18 93.57 92.65

Table 2.2.2: ResNet architectures validated with CIFAR10

[ 56 ]

% Accuracy on CIFAR10

(This book)

The following listing shows the partial code of resnet-cifar10-2.2.1.py,

which is the Keras model implementation of ResNet v1:

def resnet_v1(input_shape, depth, num_classes=10):

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

Saved successfully!

Ooh no, something went wrong!