Implementation:Kornia Kornia Image Base

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

Overview

This module defines the foundational data classes and enumerations for describing image metadata in the Kornia library, including image size, color space, pixel format, channels order, and image layout.

Description

The base module in the Kornia image package provides the core type definitions used throughout the Kornia image system. It defines five types: ImageSize (a frozen dataclass for height and width), ColorSpace (an enum with values UNKNOWN, GRAY, RGB, BGR), PixelFormat (a frozen dataclass combining color_space and bit_depth), ChannelsOrder (an enum distinguishing CHANNELS_FIRST from CHANNELS_LAST), and ImageLayout (a frozen dataclass combining image_size, channels count, and channels_order). These types are used by the Image class and other modules to enforce and communicate image structure constraints.

Usage

Import these types when constructing Image objects, when writing functions that need to reason about image layout, or when performing color space and format conversions.

Code Reference

Source Location

• Repository: Kornia
• File: kornia/image/base.py
• Lines: 1-109

Signature

@dataclass(frozen=True)
class ImageSize:
    height: int | torch.Tensor
    width: int | torch.Tensor

class ColorSpace(Enum):
    UNKNOWN = 0
    GRAY = 1
    RGB = 2
    BGR = 3

@dataclass(frozen=True)
class PixelFormat:
    color_space: ColorSpace
    bit_depth: int

class ChannelsOrder(Enum):
    CHANNELS_FIRST = 0
    CHANNELS_LAST = 1

@dataclass(frozen=True)
class ImageLayout:
    image_size: ImageSize
    channels: int
    channels_order: ChannelsOrder

Import

from kornia.image.base import ImageSize, ColorSpace, PixelFormat, ChannelsOrder, ImageLayout

I/O Contract

ImageSize

Name	Type	Required	Description
height	int or torch.Tensor	Yes	Image height in pixels.
width	int or torch.Tensor	Yes	Image width in pixels.

PixelFormat

Name	Type	Required	Description
color_space	ColorSpace	Yes	The color space (UNKNOWN, GRAY, RGB, BGR).
bit_depth	int	Yes	Number of bits per channel (e.g., 8, 16, 32).

ImageLayout

Name	Type	Required	Description
image_size	ImageSize	Yes	The spatial dimensions of the image.
channels	int	Yes	Number of image channels.
channels_order	ChannelsOrder	Yes	Whether channels are first (C, H, W) or last (H, W, C).

Usage Examples

from kornia.image.base import ImageSize, ColorSpace, PixelFormat, ChannelsOrder, ImageLayout

# Create an image size descriptor
size = ImageSize(height=480, width=640)
assert size.height == 480
assert size.width == 640

# Create a pixel format for 8-bit RGB
pixel_format = PixelFormat(color_space=ColorSpace.RGB, bit_depth=8)
assert pixel_format.color_space == ColorSpace.RGB

# Create an image layout for channels-first (PyTorch convention)
layout = ImageLayout(
    image_size=ImageSize(480, 640),
    channels=3,
    channels_order=ChannelsOrder.CHANNELS_FIRST,
)
assert layout.channels == 3
assert layout.channels_order == ChannelsOrder.CHANNELS_FIRST

Related Pages

• Principle:Kornia_Kornia_Image_Data_Model