Implementation:Kornia Kornia Generate Examples

Knowledge Sources	Kornia
Domains	Documentation, Infrastructure
Last Updated	2026-02-09 15:00 GMT

Overview

This script generates all the example images used in the Kornia documentation by programmatically applying augmentations, color transforms, filters, morphological operations, geometric transforms, and feature detectors to sample images.

Description

The generate_examples.py file in the Kornia docs directory is a standalone script that produces the static example images for the Sphinx documentation. Its main() function downloads sample images from GitHub, then iterates over dictionaries of transforms organized by Kornia submodule: augmentations (kornia.augmentation), mix augmentations, mask augmentations, augmentation containers (AugmentationSequential, PatchSequential), color transforms (kornia.color), colormaps (kornia.color.ColorMap), enhance operations (kornia.enhance), morphology operations (kornia.morphology), filters (kornia.filters), geometric transforms (kornia.geometry.transform), and local feature detectors/response functions (kornia.feature). For each transform, it generates side-by-side comparison images (original + transformed) and saves them as PNG files. Helper functions include download_tutorials_examples for fetching tutorial scripts, read_img_from_url for loading images from URLs, transparent_pad for RGBA padding, and draw_bbox_kpts for drawing bounding boxes and keypoints.

Usage

This script is executed during Sphinx documentation builds (called from docs/source/conf.py) and can also be run standalone via python docs/generate_examples.py. It produces images in docs/source/_static/img/.

Code Reference

Source Location

• Repository: Kornia
• File: docs/generate_examples.py
• Lines: 1-724

Key Signatures

def download_tutorials_examples(download_infos: dict[str, str], directory: Path): ...

def read_img_from_url(url: str, resize_to: Optional[tuple[int, int]] = None,
                      **resize_kwargs) -> torch.Tensor: ...

def transparent_pad(src: torch.Tensor, shape: tuple[int, int]) -> torch.Tensor: ...

def draw_bbox_kpts(imgs: torch.Tensor, bboxes: torch.Tensor,
                   keypoints: torch.Tensor) -> torch.Tensor: ...

def main(): ...

Import

# Typically run as a script, not imported directly
# python docs/generate_examples.py

I/O Contract

Inputs

Name	Type	Required	Description
(none)	-	-	The script fetches images from hardcoded GitHub URLs and applies transforms using Kornia APIs.

Outputs

Name	Type	Description
PNG images	Files	Generated example images saved to docs/source/_static/img/ and tutorial scripts to docs/source/_static/scripts/.

Generated Example Categories

Category	Module	Count	Description
Augmentations	kornia.augmentation	~40	RandomAffine, RandomCrop, RandomGaussianBlur, etc.
Mix Augmentations	kornia.augmentation	2	RandomMixUpV2, RandomCutMixV2
Mask Augmentations	kornia.augmentation	1	RandomTransplantation
Containers	kornia.augmentation	2	AugmentationSequential, PatchSequential
Color Transforms	kornia.color	~13	rgb_to_grayscale, rgb_to_hsv, apply_colormap, etc.
Colormaps	kornia.color	1+	ColorMapType visualization
Enhance	kornia.enhance	~13	adjust_brightness, equalize, sharpness, etc.
Morphology	kornia.morphology	7	dilation, erosion, opening, closing, etc.
Filters	kornia.filters	~14	gaussian_blur2d, sobel, canny, etc.
Geometry	kornia.geometry.transform	~16	rotate, scale, warp_affine, elastic_transform2d, etc.
Features	kornia.feature	6	harris_response, DISK, KeyNet, etc.

Usage Examples

# Run from command line to regenerate all documentation images
# cd /path/to/kornia
# python docs/generate_examples.py

# Or use specific helpers
from docs.generate_examples import read_img_from_url
img = read_img_from_url("https://raw.githubusercontent.com/kornia/data/main/panda.jpg")
# img is a torch.Tensor of shape (1, 3, H, W) normalized to [0, 1]

Related Pages

• Principle:Kornia_Kornia_Documentation_Generation