CNN architecture for image classification in PyTorch - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to define a convolutional layer in PyTorch.

PyTorch

conv1 = nn.Conv2d(in_channels=3, out_channels=16, kernel_size=[1], stride=1, padding=1)

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

A5

B7

C3

D1

Attempts:

3 left

💡 Hint

Common Mistakes

Using a kernel size too large can increase computation unnecessarily.

Using kernel size 1 reduces the receptive field.

The kernel size of 3 is commonly used in CNNs for image classification to capture local features.

2fill in blank

medium

Complete the code to add a max pooling layer after the convolution.

PyTorch

pool = nn.MaxPool2d(kernel_size=[1], stride=2)

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

A4

B3

C1

D2

Attempts:

3 left

💡 Hint

Common Mistakes

Using kernel size 1 does not reduce dimensions.

Using stride different from kernel size can cause overlapping pooling.

Max pooling with kernel size 2 and stride 2 reduces the spatial dimensions by half, which is common in CNNs.

3fill in blank

hard

Fix the error in the forward method to apply ReLU activation after convolution.

PyTorch

def forward(self, x):
    x = self.conv1(x)
    x = nn.[1]()(x)
    x = self.pool(x)
    return x

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

AReLU

BSigmoid

CSoftmax

DTanh

Attempts:

3 left

💡 Hint

Common Mistakes

Using Softmax or Sigmoid here is incorrect as they are for output layers.

Tanh can be used but ReLU is preferred for CNNs.

ReLU is the standard activation function used after convolution layers to introduce non-linearity.

4fill in blank

hard

Fill both blanks to flatten the tensor and pass it to a fully connected layer.

PyTorch

x = x.[1](x.size(0), -1)
x = self.[2](x)

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

Aview

Bfc

Creshape

Dlinear

Attempts:

3 left

💡 Hint

Common Mistakes

Using reshape instead of view can work but view is more common in PyTorch.

Calling linear directly without defining it causes errors.

Use view to flatten the tensor and then pass it to the fully connected layer named fc.

5fill in blank

hard

Fill all three blanks to define the final output layer with correct input size and activation.

PyTorch

self.fc = nn.Linear([1], [2])
output = self.fc(x)
output = nn.[3]()(output)

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

A128

B10

CLogSoftmax

DReLU

Attempts:

3 left

💡 Hint

Common Mistakes

Using ReLU in the output layer is incorrect for classification.

Incorrect input size causes shape mismatch errors.

The final linear layer maps 128 features to 10 classes, and LogSoftmax is used for classification output probabilities.