nn.MaxPool2d and nn.AvgPool2d in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output of MaxPool2d with stride and kernel size

What is the output tensor shape after applying nn.MaxPool2d with kernel_size=2 and stride=2 on a tensor of shape (1, 1, 4, 4)?

PyTorch

import torch
import torch.nn as nn

input_tensor = torch.arange(16, dtype=torch.float32).reshape(1, 1, 4, 4)
pool = nn.MaxPool2d(kernel_size=2, stride=2)
output = pool(input_tensor)
output.shape

A torch.Size([1, 1, 2, 2])

B torch.Size([1, 1, 3, 3])

C torch.Size([1, 1, 4, 4])

D torch.Size([1, 1, 1, 1])

💡 Hint

MaxPool2d reduces spatial dimensions by sliding a window and taking the max value.

MaxPool2d with kernel_size=2 and stride=2 divides each spatial dimension by 2, so 4x4 becomes 2x2.

❓ Predict Output

intermediate

Output values of AvgPool2d on a simple tensor

What is the output tensor after applying nn.AvgPool2d with kernel_size=2 and stride=2 on the following tensor? [[[[1, 2, 3, 4], [5, 6, 7, 8], [9, 10, 11, 12], [13, 14, 15, 16]]]]

PyTorch

import torch
import torch.nn as nn

input_tensor = torch.tensor([[[[1, 2, 3, 4],
                               [5, 6, 7, 8],
                               [9, 10, 11, 12],
                               [13, 14, 15, 16]]]], dtype=torch.float32)
pool = nn.AvgPool2d(kernel_size=2, stride=2)
output = pool(input_tensor)
output

A [[[[3.5, 5.5], [11.5, 13.5]]]]

B [[[[4.0, 6.0], [12.0, 14.0]]]]

C [[[[1.5, 3.5], [9.5, 11.5]]]]

D [[[[6.0, 7.0], [14.0, 15.0]]]]

💡 Hint

Average pooling computes the mean of each 2x2 block.

Each 2x2 block's average is calculated: top-left block (1+2+5+6)/4=3.5, top-right (3+4+7+8)/4=5.5, bottom-left (9+10+13+14)/4=11.5, bottom-right (11+12+15+16)/4=13.5.

❓ Model Choice

advanced

Choosing pooling layer for noise reduction

You want to reduce noise in image features while keeping the strongest signals. Which pooling layer is best?

A nn.AvgPool2d with kernel_size=1 to keep original features

B nn.MaxPool2d with stride=1 to keep all details

C nn.MaxPool2d because it keeps the strongest activation in each region

D nn.AvgPool2d because it smooths out activations by averaging

💡 Hint

Think about which pooling keeps the highest values.

MaxPool2d selects the maximum value in each region, preserving strong signals and reducing noise better than averaging.

❓ Hyperparameter

advanced

Effect of stride on output size in MaxPool2d

If you apply nn.MaxPool2d with kernel_size=3 and stride=1 on an input tensor of shape (1, 1, 7, 7), what will be the output shape?

A torch.Size([1, 1, 3, 3])

B torch.Size([1, 1, 4, 4])

C torch.Size([1, 1, 7, 7])

D torch.Size([1, 1, 5, 5])

💡 Hint

Output size = (input_size - kernel_size) / stride + 1

Output height and width = (7 - 3)/1 + 1 = 5, so output shape is (1, 1, 5, 5).

🔧 Debug

expert

Identifying error in pooling layer usage

What error will this code raise? import torch import torch.nn as nn input_tensor = torch.randn(1, 3, 32, 32) pool = nn.AvgPool2d(kernel_size=5, stride=2, padding=3) output = pool(input_tensor)

PyTorch

import torch
import torch.nn as nn

input_tensor = torch.randn(1, 3, 32, 32)
pool = nn.AvgPool2d(kernel_size=5, stride=2, padding=3)
output = pool(input_tensor)

A No error, runs successfully

B RuntimeError: padding must be less than or equal to half of kernel size

C TypeError: padding argument must be int or tuple

D ValueError: stride must be less than kernel size

💡 Hint

Check the relationship between padding and kernel size in pooling layers.

Padding cannot be larger than half the kernel size in PyTorch pooling layers, otherwise a RuntimeError occurs.