corrupt-tta 0.1.6

A modernized library for image corruptions based on ImageNet-C

corrupt_tta

corrupt_tta is a modernized Python library for applying image corruptions, based on the original ImageNet-C benchmark. It has been rewritten to use modern APIs (OpenCV, Scikit-Image, NumPy) and removes outdated dependencies like Wand (ImageMagick).

Installation

pip install corrupt_tta

Usage

import numpy as np
from PIL import Image
from corrupt_tta import corrupt

# Load an image
img = np.array(Image.open("example.jpg"))

# Apply a corruption (e.g., Gaussian Noise with severity 3)
corrupted_img = corrupt(img, severity=3, corruption_name="gaussian_noise")

# Save or display the result
Image.fromarray(corrupted_img.astype(np.uint8)).save("corrupted.jpg")

Available Corruptions

• Noise: gaussian_noise, shot_noise, impulse_noise, speckle_noise
• Blur: gaussian_blur, glass_blur, defocus_blur, motion_blur, zoom_blur
• Weather: fog, frost, snow
• Digital: contrast, brightness, saturate, jpeg_compression, pixelate, elastic_transform

Feedack

Much of this was written by an LLM agent. If you find bugs, please email: [e-mail]

Citation

If you find this useful in your research, please consider citing the original paper:

@article{hendrycks2019robustness,
  title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations},
  author={Dan Hendrycks and Thomas Dietterich},
  journal={Proceedings of the International Conference on Learning Representations},
  year={2019}
}

If you wish to reference this specific implementation, you may also cite:

@software{corrupt_tta,
  title = {corrupt_tta: Modernized ImageNet-C Corruptions Implementation},
  author = {Yash Mali},
  year = {2025},
  url = {https://github.com/YashM8/robustness}
}