CNN architecture for image classification in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output shape after convolution and pooling

Given the following PyTorch CNN layer definitions, what is the output shape after passing a batch of 16 RGB images of size 64x64 through the layers?

PyTorch

import torch
import torch.nn as nn

conv = nn.Conv2d(in_channels=3, out_channels=8, kernel_size=3, stride=1, padding=1)
pool = nn.MaxPool2d(kernel_size=2, stride=2)

x = torch.randn(16, 3, 64, 64)
x = conv(x)
x = pool(x)
output_shape = x.shape

A (16, 8, 31, 31)

B (16, 3, 32, 32)

C (16, 8, 32, 32)

D (16, 8, 64, 64)

💡 Hint

Remember that padding keeps the spatial size same before pooling, and pooling halves the height and width.

The convolution with padding=1 and kernel_size=3 keeps the height and width at 64. The max pooling with kernel_size=2 and stride=2 halves the height and width to 32. The batch size is 16 and output channels are 8.

❓ Model Choice

intermediate

Choosing the correct CNN layer sequence

Which of the following PyTorch CNN layer sequences correctly applies two convolutional layers followed by a max pooling layer?

A nn.Sequential(nn.Conv2d(3, 16, 3), nn.Conv2d(32, 16, 3), nn.MaxPool2d(2))

B nn.Sequential(nn.MaxPool2d(2), nn.Conv2d(3, 16, 3), nn.Conv2d(16, 32, 3))

C nn.Sequential(nn.Conv2d(3, 16, 3), nn.MaxPool2d(2), nn.Conv2d(16, 32, 3))

D nn.Sequential(nn.Conv2d(3, 16, 3, padding=1), nn.Conv2d(16, 32, 3, padding=1), nn.MaxPool2d(2))

💡 Hint

Check the order of layers and the input/output channels for each convolution.

Option D correctly applies Conv2d from 3 to 16 channels, then from 16 to 32 channels, then max pooling. Other options have wrong order or channel sizes.

❓ Hyperparameter

advanced

Effect of kernel size on output dimension

If a convolutional layer has input size 28x28, stride=1, padding=0, and kernel_size=5, what will be the output spatial size?

A 24x24

B 28x28

C 23x23

D 25x25

💡 Hint

Use the formula: output = (input - kernel + 2*padding)/stride + 1

Output size = (28 - 5 + 0)/1 + 1 = 24. So output is 24x24.

❓ Metrics

advanced

Interpreting CNN training accuracy and loss

During CNN training for image classification, if the training loss decreases steadily but the validation accuracy plateaus early, what does this indicate?

A The model is overfitting the training data

B The model is underfitting both training and validation data

C The model is perfectly generalized

D The training data is too small

💡 Hint

Think about what it means when training improves but validation does not.

Decreasing training loss with stagnant validation accuracy suggests the model fits training data well but fails to generalize, indicating overfitting.

🔧 Debug

expert

Identifying error in CNN forward pass

What error will this PyTorch CNN forward method raise when given input of shape (10, 3, 32, 32)?

PyTorch

import torch
import torch.nn as nn

class SimpleCNN(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(3, 16, 3)
        self.pool = nn.MaxPool2d(2)
        self.fc1 = nn.Linear(16 * 15 * 15, 10)

    def forward(self, x):
        x = self.pool(self.conv1(x))
        x = x.view(-1, 16 * 15 * 15)
        x = self.fc1(x)
        return x

model = SimpleCNN()
input_tensor = torch.randn(10, 3, 32, 32)
output = model(input_tensor)

A ValueError: input tensor has wrong number of channels

B RuntimeError: shape mismatch in view

C TypeError: conv1d expected 3D input

D No error, output shape is (10, 10)

💡 Hint

Check the size after conv and pooling before reshaping with view.

After conv1 with kernel=3 and default stride=1, input 32x32 becomes 30x30. MaxPool2d(2) halves it to 15x15. So x shape is (10,16,15,15). The view tries to reshape to (10, 16*16*16=4096), but actual features are 16*15*15=3600, causing RuntimeError.