Implementation:Kornia Kornia Augmentation Transforms 2D

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

Overview

Concrete tool for defining differentiable 2D image augmentation transforms provided by Kornia's augmentation module.

Description

Kornia provides a rich set of 2D augmentation transforms inheriting from _BasicAugmentationBase. All transforms are nn.Module instances with probabilistic application, batch support, and transformation matrix tracking for invertibility.

Key transforms include:

RandomAffine -- rotation, translation, scale, and shear in a single transform
ColorJiggle -- random perturbation of brightness, contrast, saturation, and hue
RandomPerspective -- perspective distortion with configurable distortion scale
RandomBrightness -- dedicated brightness adjustment

Each transform accepts a probability parameter p that controls how often the transform is applied during training. When a transform is not applied (based on the probability check), the input is passed through unchanged.

Usage

Import specific transforms when building augmentation pipelines for training. Compose them using AugmentationSequential for multi-data-key support (images, masks, bounding boxes, keypoints).

Code Reference

Source Location

Repository: kornia
File: kornia/augmentation/_2d/ (individual transform files), base at kornia/augmentation/base.py
Lines: base.py L51-571

Signature

# Base class
class _BasicAugmentationBase(nn.Module):
    def __init__(
        self,
        p: float = 0.5,
        p_batch: float = 1.0,
        same_on_batch: bool = False,
        keepdim: bool = False,
    ) -> None

# RandomAffine
RandomAffine(
    degrees,
    translate=None,
    scale=None,
    shear=None,
    p=0.5,
    same_on_batch=False,
    keepdim=False,
)

# ColorJiggle
ColorJiggle(
    brightness=0.,
    contrast=0.,
    saturation=0.,
    hue=0.,
    p=1.0,
    same_on_batch=False,
    keepdim=False,
)

Import

from kornia.augmentation import RandomAffine, ColorJiggle, RandomPerspective, RandomBrightness

I/O Contract

Inputs

Name	Type	Required	Description
degrees	float or tuple	Yes (RandomAffine)	Range of rotation degrees
translate	tuple	No	Range of horizontal and vertical translation as fraction of image size
scale	tuple	No	Range of scaling factor
shear	float or tuple	No	Range of shear degrees
brightness	float or tuple	No (ColorJiggle)	Brightness jitter range
contrast	float or tuple	No (ColorJiggle)	Contrast jitter range
saturation	float or tuple	No (ColorJiggle)	Saturation jitter range
hue	float or tuple	No (ColorJiggle)	Hue jitter range
p	float	No	Probability of applying the transform; default varies by transform

Outputs

Name	Type	Description
return	nn.Module	Instantiated transform module accepting (B, C, H, W) tensors, producing augmented (B, C, H, W) tensors

Usage Examples

RandomAffine with Rotation and Scale

import torch
from kornia.augmentation import RandomAffine

# Create a RandomAffine transform with rotation [-30, 30] degrees and scale [0.8, 1.2]
aug = RandomAffine(degrees=(-30, 30), scale=(0.8, 1.2), p=1.0)

# Apply to a batch of images
images = torch.randn(4, 3, 256, 256)  # (B, C, H, W)
augmented = aug(images)
print(augmented.shape)  # torch.Size([4, 3, 256, 256])

ColorJiggle for Photometric Augmentation

import torch
from kornia.augmentation import ColorJiggle

# Create a ColorJiggle transform
aug = ColorJiggle(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.1, p=1.0)

# Apply to a batch of images
images = torch.randn(4, 3, 256, 256)
augmented = aug(images)
print(augmented.shape)  # torch.Size([4, 3, 256, 256])

Related Pages

Implements Principle

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment