Implementation:Kornia Kornia Sepia Filter

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

Overview

This module applies a sepia tone filter to RGB image tensors using a fixed color transformation matrix.

Description

sepia.py is a module in the Kornia library's color subpackage that implements a sepia tone filter for RGB images. The sepia_from_rgb function applies a standard sepia transformation matrix to the RGB channels: R_out = 0.393*R + 0.769*G + 0.189*B, G_out = 0.349*R + 0.686*G + 0.168*B, B_out = 0.272*R + 0.534*G + 0.131*B. An optional rescaling step normalizes the output so that the maximum pixel value is 1.0 (or 255 for integer inputs). The Sepia nn.Module class wraps this function with configurable rescale and epsilon parameters.

Usage

Import this module when you need to apply a vintage sepia tone effect to images in a differentiable pipeline, for example in image style transfer, augmentation, or artistic filter applications.

Code Reference

Source Location

• Repository: Kornia
• File: kornia/color/sepia.py
• Lines: 1-97

Signature

def sepia_from_rgb(input: torch.Tensor, rescale: bool = True, eps: float = 1e-6) -> torch.Tensor: ...

class Sepia(nn.Module):
    def __init__(self, rescale: bool = True, eps: float = 1e-6) -> None: ...
    def forward(self, input: torch.Tensor) -> torch.Tensor: ...

Import

from kornia.color import sepia_from_rgb, Sepia

I/O Contract

Inputs

Name	Type	Required	Description
input	torch.Tensor	Yes	RGB image with shape (*, 3, H, W).
rescale	bool	No	If True, rescale the output so the max value is 1.0. Default: True.
eps	float	No	Epsilon for numerical stability during rescaling. Default: 1e-6.

Outputs

Name	Type	Description
return	torch.Tensor	Sepia-toned image with shape (*, 3, H, W). If rescale=True, values are normalized to max 1.0. If rescale=False, values may exceed 1.0.

Usage Examples

Basic Usage

import torch
from kornia.color import sepia_from_rgb, Sepia

# Apply sepia filter with rescaling (default)
rgb = torch.rand(1, 3, 256, 256)
sepia_img = sepia_from_rgb(rgb)
print(sepia_img.shape)  # torch.Size([1, 3, 256, 256])
print(sepia_img.max())  # approximately 1.0 due to rescaling

# Apply sepia filter without rescaling
sepia_raw = sepia_from_rgb(rgb, rescale=False)
print(sepia_raw.max())  # may exceed 1.0

# Using the nn.Module wrapper
sepia_module = Sepia(rescale=True, eps=1e-6)
sepia_out = sepia_module(rgb)
print(sepia_out.shape)  # torch.Size([1, 3, 256, 256])

Related Pages

• Principle:Kornia_Kornia_Sepia_Tone_Effect