Augmentation

This page outlines the unit test coverage for the augmentation.py module, which applies configurable image augmentations to labeled datasets using Albumentations. It ensures the generation of augmented images and updated label JSONs while preserving bounding box structure.

Coverage Overview

This suite tests the following components:

  • build_augmentation_transform: Builds the augmentation pipeline from a configuration dictionary.

  • augment_images: Applies augmentations to matched image-label pairs.

  • augment_dataset: Full pipeline integration test using a real image and JSON file.

Fixtures

sample_config

Provides a valid augmentation configuration dictionary with all probabilities set to 1.0 to force all transforms to apply. Includes:

  • Horizontal flip

  • Brightness/contrast

  • Hue/saturation

  • Gaussian blur and noise

  • Grayscale

  • Rotation

temp_image_and_json

Creates a temporary test image and its corresponding JSON label in a valid format:

  • Bounding box: [10, 10, 50, 50]

  • Class: "FireBSI"

  • Confidence: 0.9

Function Tests

build_augmentation_transform

  • test_build_augmentation_transform:
    Validates that the transform pipeline builds successfully and is callable.

augment_images

  • test_augment_images_creates_augmented_files:
    Verifies that the correct number of augmented images and labels are created based on num_augmentations.
    Also checks that each output JSON contains a valid predictions list.

augment_dataset

  • test_augment_dataset_integration:
    Integration test that simulates a full run of the augment_dataset() pipeline.

    It:

    • Copies the image to an input directory

    • Creates a valid labeled directory containing its JSON label

    • Runs the pipeline

    • Verifies that augmented images and labels are created in the output directory

Summary

This page outlines a focused test suite ensuring:

  • Augmentation configurations are valid and interpretable

  • Image-label pairs are augmented correctly and written to disk

  • End-to-end augmentation logic functions as expected in a simulated pipeline environment

These tests confirm that the augmentation module integrates safely and predictably with the rest of the AutoML pipeline.