facebookresearch/AugLy

A data augmentations library for audio, image, text, and video.

What it solves

AugLy addresses the need for diverse and realistic data augmentation to improve the robustness of AI models. It specifically targets the "internet user" style of modifications—such as turning images into memes or adding social media screenshots—that often cause models to fail in real-world scenarios like hate speech detection or copyright infringement tracking.

How it works

The library provides over 100 augmentations across four modalities: audio, image, text, and video. It uses a combination of function-based and class-based transforms and composition operators to modify data. Users can also track metadata about the transforms applied, including the intensity of the change.

Who it’s for

It is designed for machine learning engineers and researchers working on model training or robustness evaluation, particularly those dealing with content moderation, copy detection, and copyright infringement.

Highlights

  • Multimodal Support: Covers audio, image, text, and video in a single library.
  • Internet-Realistic Augmentations: Includes specific tools to create memes, overlay emojis, and simulate social media screenshots.
  • Robustness Testing: Useful for identifying robustness gaps in existing models.
  • Extensible Architecture: Offers both function-based and class-based transforms with metadata tracking.

Related

  • Project
    iver56/audiomentations

    A Python library for audio data augmentation that provides a wide variety of transforms to make audio deep learning models more robust for real-world use.

  • Project
    makcedward/nlpaug

    A Python library for augmenting text, audio, and spectrogram data to generate synthetic training sets for machine learning models.

  • Dispatch
    Hugging Face TextImage Augmentation pipeline release

    Hugging Face and Albumentations AI released a TextImage Augmentation pipeline that jointly modifies document images and their text, enabling realistic synthetic data generation and robust fine‑tuning of vision‑language models on limited document datasets.

  • Project
    google-research/augmix

    AugMix is a data processing technique that mixes augmented images and enforces consistent embeddings to improve the robustness and and uncertainty calibration of image classification models.

  • Project
    lhotse-speech/lhotse

    A Python library for flexible multimodal data preparation that optimizes the loading and manipulation of speech, audio, video, and image data for ML training.