nn.Conv2d layers in PyTorch - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output shape after Conv2d layer

What is the output shape of the tensor after applying this Conv2d layer?

Input tensor shape: (batch_size=1, channels=3, height=32, width=32)
Layer: nn.Conv2d(in_channels=3, out_channels=6, kernel_size=5, stride=1, padding=0)

PyTorch

import torch
import torch.nn as nn

conv = nn.Conv2d(in_channels=3, out_channels=6, kernel_size=5, stride=1, padding=0)
input_tensor = torch.randn(1, 3, 32, 32)
output = conv(input_tensor)
print(output.shape)

A (1, 6, 27, 27)

B (1, 3, 28, 28)

C (1, 6, 28, 28)

D (1, 6, 32, 32)

💡 Hint

Remember the formula for output size: (W - K + 2P) / S + 1

The output height and width are calculated as (32 - 5 + 0) / 1 + 1 = 28. The output channels are 6 as set by out_channels.

❓ Model Choice

intermediate

Choosing Conv2d parameters for same input-output size

You want a Conv2d layer that keeps the input height and width the same after convolution.
Input shape: (batch_size=1, channels=3, height=28, width=28)
Which Conv2d layer parameters will achieve this?

A nn.Conv2d(3, 10, kernel_size=3, stride=1, padding=1)

B nn.Conv2d(3, 10, kernel_size=5, stride=1, padding=0)

C nn.Conv2d(3, 10, kernel_size=3, stride=2, padding=1)

D nn.Conv2d(3, 10, kernel_size=1, stride=1, padding=0)

💡 Hint

Padding can help keep the size same if chosen correctly.

Using kernel_size=3, stride=1, and padding=1 keeps the height and width unchanged because (28 - 3 + 2*1)/1 + 1 = 28.

❓ Hyperparameter

advanced

Effect of stride on Conv2d output size

Given an input tensor of shape (1, 3, 64, 64), which Conv2d layer will produce an output with height and width equal to 16?

A nn.Conv2d(3, 8, kernel_size=3, stride=4, padding=0)

B nn.Conv2d(3, 8, kernel_size=5, stride=2, padding=0)

C nn.Conv2d(3, 8, kernel_size=3, stride=3, padding=1)

D nn.Conv2d(3, 8, kernel_size=7, stride=4, padding=0)

💡 Hint

Use the formula: output = (W - K + 2P)/S + 1 and solve for output=16.

For option A: (64 - 3 + 0)/4 + 1 = 15.25 + 1 = 16.25, which floors to 16 in PyTorch. Others produce different sizes.

🔧 Debug

advanced

Identifying error in Conv2d layer usage

What error will this code raise when run?

conv = nn.Conv2d(in_channels=3, out_channels=6, kernel_size=3)
input_tensor = torch.randn(1, 1, 28, 28)
output = conv(input_tensor)

PyTorch

import torch
import torch.nn as nn
conv = nn.Conv2d(in_channels=3, out_channels=6, kernel_size=3)
input_tensor = torch.randn(1, 1, 28, 28)
output = conv(input_tensor)

A TypeError: Conv2d() missing required positional argument 'stride'

B RuntimeError: Given groups=1, weight of size [6, 3, 3, 3], expected input[1, 1, 28, 28] to have 3 channels, but got 1 channels instead

C ValueError: kernel_size must be odd

D No error, runs successfully

💡 Hint

Check if input channels match the Conv2d in_channels parameter.

The input tensor has 1 channel but Conv2d expects 3 channels, causing a runtime error.

🧠 Conceptual

expert

Understanding dilation effect in Conv2d

How does increasing the dilation parameter in nn.Conv2d affect the receptive field and output size?

Choose the correct statement.

A Dilation has no effect on receptive field or output size, only on computation speed

B Increasing dilation decreases the receptive field and increases output size

C Increasing dilation increases output size but keeps receptive field constant

D Increasing dilation increases the receptive field without changing the kernel size, and reduces the output size if padding is not adjusted

💡 Hint

Think of dilation as spacing out kernel elements, making the filter cover a larger area.

Dilation spaces out kernel elements, increasing receptive field. Without adjusting padding, output size shrinks because effective kernel size grows.