Implementation:Kornia Kornia Sobel Operator

Knowledge Sources	Kornia
Domains	Vision Filters
Last Updated	2026-02-09 15:00 GMT

Overview

Concrete tool for computing Sobel gradient edge magnitude provided by Kornia's filters module.

Description

The Sobel class computes the gradient magnitude per channel using Sobel operators. It applies horizontal and vertical 3x3 Sobel kernels, computes the magnitude sqrt(gx^2 + gy^2), and optionally normalizes the kernel. The output preserves the per-channel structure of the input.

The related Laplacian class computes second-order derivative edges using a configurable kernel size.

Usage

Import for gradient-based edge detection or as a building block for custom edge detection pipelines.

Code Reference

Source Location

• Repository: kornia
• File: kornia/filters/sobel.py (Sobel L251-280), kornia/filters/laplacian.py (Laplacian L66-109)

Signature

Sobel:

class Sobel(nn.Module):
    def __init__(self, normalized: bool = True, eps: float = 1e-6) -> None

    def forward(self, input: torch.Tensor) -> torch.Tensor

Laplacian:

class Laplacian(nn.Module):
    def __init__(
        self,
        kernel_size: tuple[int, int] | int,
        border_type: str = "reflect",
        normalized: bool = True,
    ) -> None

    def forward(self, input: torch.Tensor) -> torch.Tensor

Import

from kornia.filters import Sobel, Laplacian

I/O Contract

Inputs

Sobel constructor parameters:

Name	Type	Required	Description
normalized	bool	No	Normalize the Sobel kernel (default True)
eps	float	No	Regularization constant (default 1e-6)

Sobel forward parameters:

Name	Type	Required	Description
input	torch.Tensor	Yes	Input image tensor of shape (B, C, H, W)

Laplacian constructor parameters:

Name	Type	Required	Description
kernel_size	int	Yes	Size of the Laplacian kernel
border_type	str	No	Padding mode (default "reflect")
normalized	bool	No	Normalize the kernel (default True)

Laplacian forward parameters:

Name	Type	Required	Description
input	torch.Tensor	Yes	Input image tensor of shape (B, C, H, W)

Outputs

Sobel:

Name	Type	Description
output	torch.Tensor	Gradient magnitude of shape (B, C, H, W)

Laplacian:

Name	Type	Description
output	torch.Tensor	Laplacian response of shape (B, C, H, W)

Usage Examples

Sobel Edge Detection

import torch
from kornia.filters import Sobel

# Create Sobel operator
sobel = Sobel(normalized=True)

# Compute gradient magnitude
img = torch.rand(1, 3, 256, 256)
edges = sobel(img)  # (1, 3, 256, 256)

Laplacian Edge Detection

import torch
from kornia.filters import Laplacian

# Create Laplacian operator
laplacian = Laplacian(kernel_size=3)

# Compute Laplacian edges
img = torch.rand(1, 1, 256, 256)
edges = laplacian(img)  # (1, 1, 256, 256)

Related Pages

Implements Principle

• Principle:Kornia_Kornia_Gradient_Edge_Detection
• Environment:Kornia_Kornia_PyTorch_Python_Environment