Kernel size, stride, padding in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output shape after convolution with stride and padding

Given a 1-channel input tensor of shape (1, 1, 10, 10), a convolutional layer with kernel size 3, stride 2, and padding 1 is applied. What is the shape of the output tensor?

PyTorch

import torch
import torch.nn as nn

input_tensor = torch.randn(1, 1, 10, 10)
conv = nn.Conv2d(in_channels=1, out_channels=1, kernel_size=3, stride=2, padding=1)
output = conv(input_tensor)
output.shape

A (1, 1, 6, 6)

B (1, 1, 4, 4)

C (1, 1, 7, 7)

D (1, 1, 5, 5)

💡 Hint

Recall the formula for output size: floor((input + 2*padding - kernel_size)/stride) + 1

The output height and width are calculated as floor((10 + 2*1 - 3)/2) + 1 = floor(9/2) + 1 = 4 + 1 = 5. So output shape is (1, 1, 5, 5).

🧠 Conceptual

intermediate

Effect of stride on convolution output size

Which statement correctly describes the effect of increasing the stride in a convolutional layer?

A Increasing stride has no effect on output spatial dimensions.

B Increasing stride increases the output spatial dimensions.

C Increasing stride decreases the output spatial dimensions.

D Increasing stride increases the number of output channels.

💡 Hint

Think about how stride controls the step size when sliding the kernel.

Stride controls how far the kernel moves each step. Larger stride means fewer positions, so output size decreases.

❓ Metrics

advanced

Padding effect on output size with kernel size 5

A convolutional layer uses kernel size 5 and stride 1 on an input of size 28x28. Which padding value will keep the output size the same as the input size?

A Padding = 2

B Padding = 1

C Padding = 0

D Padding = 3

💡 Hint

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

To keep output size same as input (28), solve 28 = floor((28 + 2*padding - 5)/1) + 1. This gives padding = 2.

🔧 Debug

advanced

Identify the error in convolution output size calculation

A user calculates the output size of a convolution as output = (input_size - kernel_size + 2*padding) / stride + 1 without flooring the division. What problem can this cause?

A Output size will always be larger than input size.

B Output size may be a float, which is invalid for tensor dimensions.

C Padding will be ignored during convolution.

D Stride will be treated as 1 regardless of value.

💡 Hint

Tensor shapes must be integers.

Output size must be an integer because tensor dimensions cannot be fractional. Not flooring leads to float output size, causing errors.

❓ Model Choice

expert

Choosing padding and stride for specific output size

You want a convolutional layer with kernel size 4 on a 32x32 input to produce an output of size 15x15. Which combination of stride and padding achieves this?

A Stride = 2, Padding = 0

B Stride = 3, Padding = 0

C Stride = 1, Padding = 1

D Stride = 2, Padding = 1

💡 Hint

Use output size formula and solve for stride and padding.

Using output = floor((32 + 2*padding - kernel_size)/stride) + 1, with stride=2 and padding=0: floor((32 + 0 - 4)/2) + 1 = floor(28/2) + 1 = 14 + 1 = 15.