ResNet and skip connections in Computer Vision - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to add a skip connection by summing the input tensor with the output tensor.

Computer Vision

def residual_block(x):
    out = some_layer(x)
    out = some_activation(out)
    return out [1] x

Drag options to blanks, or click blank then click option'

A*

B-

C+

D/

Attempts:

3 left

💡 Hint

Common Mistakes

Using subtraction or multiplication instead of addition.

Forgetting to add the input tensor back.

In ResNet, the skip connection adds the input tensor to the output tensor using the '+' operator.

2fill in blank

medium

Complete the code to apply batch normalization after the convolution layer in the residual block.

Computer Vision

def residual_block(x):
    out = Conv2D(64, (3, 3), padding='same')(x)
    out = [1](out)
    out = Activation('relu')(out)
    return out + x

Drag options to blanks, or click blank then click option'

ABatchNormalization

BDropout

CMaxPooling2D

DFlatten

Attempts:

3 left

💡 Hint

Common Mistakes

Using Dropout or pooling layers instead of normalization.

Placing BatchNormalization after activation instead of before.

BatchNormalization is used to normalize the output of the convolution layer before activation.

3fill in blank

hard

Fix the error in the code to correctly implement a skip connection when the input and output shapes differ.

Computer Vision

def residual_block(x):
    out = Conv2D(128, (3, 3), strides=2, padding='same')(x)
    out = BatchNormalization()(out)
    out = Activation('relu')(out)
    shortcut = [1](x)
    return out + shortcut

Drag options to blanks, or click blank then click option'

AMaxPooling2D(pool_size=(2, 2))

BConv2D(128, (1, 1), strides=2, padding='same')

CDense(128)

DFlatten()

Attempts:

3 left

💡 Hint

Common Mistakes

Using pooling or dense layers which change shape incorrectly.

Not adjusting the shortcut shape causing shape mismatch errors.

When input and output shapes differ, a 1x1 convolution with matching strides adjusts the shortcut to the correct shape.

4fill in blank

hard

Fill both blanks to create a residual block that applies two convolution layers with batch normalization and adds the skip connection.

Computer Vision

def residual_block(x):
    out = Conv2D(64, (3, 3), padding='same')(x)
    out = [1](out)
    out = Activation('relu')(out)
    out = Conv2D(64, (3, 3), padding='same')(out)
    out = [2](out)
    return out + x

Drag options to blanks, or click blank then click option'

ABatchNormalization

BDropout

CMaxPooling2D

DFlatten

Attempts:

3 left

💡 Hint

Common Mistakes

Using Dropout or pooling layers instead of BatchNormalization.

Skipping normalization after the second convolution.

BatchNormalization is applied after each convolution layer before activation or addition.

5fill in blank

hard

Fill all three blanks to define a ResNet block that adjusts the shortcut path and applies two convolution layers with batch normalization.

Computer Vision

def resnet_block(x):
    shortcut = [1](x)
    out = Conv2D(128, (3, 3), strides=2, padding='same')(x)
    out = [2](out)
    out = Activation('relu')(out)
    out = Conv2D(128, (3, 3), padding='same')(out)
    out = [3](out)
    return out + shortcut

Drag options to blanks, or click blank then click option'

AConv2D(128, (1, 1), strides=2, padding='same')

BBatchNormalization

DMaxPooling2D(pool_size=(2, 2))

Attempts:

3 left

💡 Hint

Common Mistakes

Using pooling instead of convolution for shortcut.

Skipping batch normalization layers.

The shortcut uses a 1x1 convolution to match shape, and batch normalization is applied after each convolution layer.