nn.Conv2d layers in PyTorch - Cheat Sheet & Quick Revision

Recall & Review

beginner

What does nn.Conv2d layer do in a neural network?

It applies a set of filters (kernels) to 2D input data (like images) to extract features such as edges or textures by sliding the filters over the input.

beginner

What are the main parameters of nn.Conv2d?

The main parameters are: in_channels (input depth), out_channels (number of filters), kernel_size (filter size), stride (step size for sliding), padding (border added), and dilation (spacing inside the kernel).

intermediate

How does padding affect the output size of nn.Conv2d?

Padding adds pixels around the input edges, allowing the filter to cover border areas. This can keep the output size the same as input or control shrinking.

intermediate

What is the effect of stride in nn.Conv2d?

Stride controls how far the filter moves each step. A stride of 1 moves one pixel at a time, producing larger outputs. Larger strides skip pixels, reducing output size.

beginner

Why do we use multiple filters (out_channels) in nn.Conv2d?

Multiple filters let the network learn different features from the input, like edges, colors, or shapes, improving the model's ability to understand complex patterns.

What does the kernel_size parameter in nn.Conv2d specify?

A The number of filters used

B The size of the filter applied to the input

C The step size of the filter movement

D The amount of padding added

kernel_size defines the height and width of the filter sliding over the input.

If stride is set to 2 in nn.Conv2d, what happens to the output size compared to stride 1?

A Output size becomes zero

B Output size doubles

C Output size stays the same

D Output size halves approximately

A stride of 2 moves the filter two pixels at a time, reducing output size roughly by half.

What is the role of padding in nn.Conv2d?

A To add borders so output size can be controlled

B To speed up training

C To change the input channels

D To increase the number of filters

Padding adds pixels around input edges to control output size and preserve border information.

What does out_channels parameter control in nn.Conv2d?

A Number of filters applied

B Number of input channels

C Size of the input image

D Stride length

out_channels sets how many filters the layer uses, determining output depth.

Which of these is NOT a typical use of nn.Conv2d?

A Extracting image features

B Reducing image size

C Classifying text documents directly

D Detecting edges in images

nn.Conv2d is designed for 2D data like images, not for direct text classification.

Explain how the parameters kernel_size, stride, and padding affect the output size of an nn.Conv2d layer.

Think about how the filter moves and covers the input image.

Describe why multiple filters (out_channels) are used in nn.Conv2d and how they help the model learn.

Imagine looking at an image with different colored glasses to see different details.