#0019

Convolution Output Size

MediumML18 min15 XP

Problem

Given input_size, kernel_size, padding, and stride, return the output size for one spatial dimension using floor((input + 2*padding - kernel) / stride) + 1.

Why This Matters

CNN debugging often starts with shape debugging. If you can predict output size, model architecture becomes much easier to reason about.

Function Signature

def conv_output_size(input_size, kernel_size, padding, stride):

Examples

Example 1

Inputinput 32, kernel 3, padding 1, stride 1

Output32

Padding 1 keeps a 3x3 convolution from shrinking the image when stride is 1.

Example 2

Inputinput 28, kernel 5, padding 0, stride 2

Output12

The kernel moves in steps of 2, so the output is smaller.

Constraints

All inputs are positive integers except padding, which may be zero.
Use integer floor division.
Return the output size for one dimension only.

CodePython

def conv_output_size(input_size, kernel_size, padding, stride):
    """
    Input example: input 32, kernel 3, padding 1, stride 1
    Return: 32
    Return the value from the function so the tests can compare it. Do not print unless the prompt asks.
    """
    # write your code here
    return input_size

def conv_output_size(input_size, kernel_size, padding, stride):
    """
    Input example: input 32, kernel 3, padding 1, stride 1
    Return: 32
    Return the value from the function so the tests can compare it. Do not print unless the prompt asks.
    """
    # write your code here
    return input_size

Visible browser tests run here when available.

Testcases2 visible / 4 hidden categories

Same-size 3x3 convolution

Input32, 3, 1, 1

Expected32

This is the common same-padding case for odd kernels.

Stride shrinks output

Input28, 5, 0, 2

Expected12

Floor division handles partial final placements.

Hidden Test Categories

Kernel size 1Large paddingStride 3Input barely larger than kernel