Data

Vision data utilities for image processing and dataset management.

This module provides utilities for working with image data in machine learning pipelines. It includes classes and functions for loading, managing, and processing image datasets with support for various image formats and transformations.

The module is designed to work seamlessly with PIL (Python Imaging Library) and integrates with the broader cmn_ai data processing pipeline.

Functions:

• None –

Classes:

• ImageList : Image dataset container for managing collections of image files –

Constants

IMAGE_EXTENSIONS : list of str List of supported image file extensions derived from mimetypes.

Notes

This module extends the base data utilities to provide vision-specific functionality for image processing workflows.

Examples:

>>> from cmn_ai.vision.data import ImageList
>>> # Create image dataset from directory
>>> images = ImageList.from_files('./data/images')
>>> # Access first image
>>> img = images[0]  # Returns PIL.Image object
>>> print(img.size)  # (width, height)

ImageList

Bases: ItemList

Image dataset container for managing collections of image files.

A specialized ItemList subclass for handling image datasets. It provides functionality to load and manage collections of image files with support for various image formats, transformations, and file discovery.

Attributes:

• path (Path) –

  Base path where the image files are located.

• tfms ((callable, optional)) –

  Transformations to apply to images when retrieved.

• data (list[Path]) –

  List of image file paths.

Notes

This class extends ItemList to provide image-specific functionality. Images are loaded using PIL (Python Imaging Library) when accessed.

Examples:

>>> # Create ImageList from a directory
>>> image_list = ImageList.from_files('./data/images')
>>> # Access first image
>>> img = image_list[0]  # Returns PIL.Image object
>>> # Create with transformations
>>> from torchvision import transforms
>>> tfms = transforms.Compose([transforms.Resize((224, 224))])
>>> image_list = ImageList.from_files('./data', tfms=tfms)

from_files(path, extensions=IMAGE_EXTENSIONS, include=None, recurse=True, tfms=None) classmethod

Create an ImageList from files in a directory.

Factory method to create an ImageList by discovering image files in the specified directory. Supports various image formats and recursive directory traversal.

Parameters:

• path (str or Path) –

  Path for the root directory to search for image files.

• extensions (str or iterable of str, default: IMAGE_EXTENSIONS ) –

  File extensions to include. Defaults to all supported image formats.

• include (iterable of str, default: None ) –

  Top-level directory(ies) under path to use for searching files. If None, searches all directories.

• recurse (bool, default: True ) –

  Whether to search subdirectories recursively.

• tfms (callable, default: None ) –

  Transformations to apply to images before returning them.

Returns:

• ImageList –

  New ImageList instance containing discovered image files.

Examples:

>>> # Create from directory with default settings
>>> images = ImageList.from_files('./data/images')
>>> # Create with specific extensions
>>> images = ImageList.from_files('./data', extensions=['.jpg', '.png'])
>>> # Create without recursion
>>> images = ImageList.from_files('./data', recurse=False)

get(item)

Load and return an image from file path.

Opens an image file using PIL and returns the PIL Image object. This method is called automatically when accessing items from the ImageList.

Parameters:

• item (Path) –

  File path to the image to load.

Returns:

• Image –

  Loaded image object.

Notes

The image is loaded in its original format. If transformations are specified in the ImageList, they will be applied after loading.

Examples:

>>> image_list = ImageList.from_files('./data')
>>> img = image_list.get(Path('./data/image.jpg'))
>>> print(img.size)  # (width, height)