Implementation:Kornia Kornia Canny Edge Detector

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

Overview

Concrete tool for differentiable multi-stage Canny edge detection provided by Kornia's filters module.

Description

The Canny class implements the full Canny edge detection pipeline as a differentiable nn.Module. It performs Gaussian smoothing, Sobel gradient computation, non-maximum suppression, double thresholding, and optional hysteresis. The module returns both the gradient magnitude map and the thresholded binary edge map. Thresholds operate in the [0, 1] range on normalized images. The implementation is fully differentiable for use in training loops.

Usage

Import when you need classical edge detection with configurable sensitivity. Wrap in AugmentationSequential or use standalone.

Code Reference

Source Location

• Repository: kornia
• File: kornia/filters/canny.py
• Lines: L164-242

Signature

class Canny(nn.Module):
    def __init__(
        self,
        low_threshold: float = 0.1,
        high_threshold: float = 0.2,
        kernel_size: tuple[int, int] | int = (5, 5),
        sigma: tuple[float, float] | torch.Tensor = (1, 1),
        hysteresis: bool = True,
        eps: float = 1e-6,
    ) -> None

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

Import

from kornia.filters import Canny

I/O Contract

Inputs

Constructor parameters:

Name	Type	Required	Description
low_threshold	float	No	Lower hysteresis threshold (default 0.1)
high_threshold	float	No	Upper hysteresis threshold (default 0.2)
kernel_size	int	No	Gaussian kernel size (default (5, 5))
sigma	torch.Tensor	No	Gaussian sigma (default (1, 1))
hysteresis	bool	No	Enable hysteresis edge tracking (default True)
eps	float	No	Regularization constant (default 1e-6)

Forward parameters:

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

Outputs

Name	Type	Description
magnitude	torch.Tensor	Gradient magnitude map of shape (B, 1, H, W)
edges	torch.Tensor	Binary edge map of shape (B, 1, H, W)

Usage Examples

Basic Example

import torch
from kornia.filters import Canny
from kornia.io import load_image
from kornia.utils import tensor_to_image

# Load image
img = load_image("photo.jpg", device="cuda")[None]  # (1, 3, H, W)

# Create Canny detector with custom thresholds
canny = Canny(low_threshold=0.1, high_threshold=0.3, kernel_size=5)

# Detect edges
magnitude, edges = canny(img)

# Visualize
import matplotlib.pyplot as plt
plt.imshow(tensor_to_image(edges[0]), cmap="gray")
plt.show()

Related Pages

Implements Principle

• Principle:Kornia_Kornia_Edge_Detection
• Environment:Kornia_Kornia_PyTorch_Python_Environment