facebookresearch/AugLy
AugLy
AugLy is a multimodal data augmentation library and machine learning dataset augmentor. It provides a system for generating synthetic variations of training data across audio, image, text, and video datasets to increase sample diversity and improve model robustness.
The library functions as a multimedia noise simulator, specifically designed to mimic real-world user captures by overlaying social media templates and internet artifacts onto media. It includes a data provenance tracker to record the specific transformations and intensity levels applied to each piece of augmented data.
The tool covers a broad range of dataset expansion capabilities, including linguistic transformations for text, temporal and visual transformations for video, and sonic transformations for audio.
Features
- Dataset Augmentors - A tool for generating synthetic variations of training data to improve the robustness and generalization of AI models.
- Multimedia Dataset Augmentation - Generates synthetic variations of training data across audio, image, text, and video datasets to increase sample diversity.
- Training Data Augmentation - Applies sonic transformations to audio datasets to increase variety and robustness for machine learning models.
- Image Augmentation - Applies visual transformations to image datasets to increase sample diversity and improve model robustness.
- Audio Dataset Preprocessing - Applies transformations to audio files to create more varied training samples for sound recognition or processing models.
- Image Dataset Expansion Frameworks - Applies visual transformations to image datasets to increase sample variety and help models generalize better.
- Machine Learning Data Augmentation - Increases the diversity of audio, image, text, and video training sets to improve the robustness of machine learning models.
- Multimodal Data Augmentation - Apply transformations to diverse datasets using logic-based rules to increase data variety.
- Text Data Augmentation - Applies linguistic transformations to text datasets to increase sample diversity and improve model robustness.
- Unified Multimodal APIs - Offers a unified API interface to apply transformations across diverse modalities including text, audio, and video.
- Video Data Augmentation - Applies temporal and visual transformations to video datasets to expand the diversity of training samples.
- Video Data Augmentations - Provides temporal and visual transformations for video clips to expand the diversity of machine learning training samples.
- Social Media Noise Simulations - Simulates real-world user captures by overlaying social media templates and internet artifacts onto images and videos.
- Augmentation Provenance Tracking - Includes a data provenance tracker to record the specific transformations and intensity levels applied to each augmented data point.
- Internet Artifact Simulation - Overlays text or social elements onto media to mimic real-world user behavior and internet artifacts for robustness testing.
- Social Media Capture Simulation - Overlays images or videos onto feed templates to mimic how users capture content in real-world social media environments.
- Social Media Noise Simulation - Adds overlays and templates to media to mimic real-world social media screenshots for testing model robustness against noise.
- UI Template Overlays - Combines media samples with predefined social media layout templates to simulate realistic user interface captures.
- Data Provenance Recorders - Records the specific transformations and intensity levels applied to augmented data to maintain a clear history of provenance.
- Computer Vision - Multimodal augmentation library.
- Deep Learning and Computer Vision - Data augmentation library for multiple media types.
- Data Wrangling - Multi-modal data augmentation library for audio, image, and text.
- Image Augmentation - Augmentation library for audio, image, text, and video data.
Star history
How this analysis was created: This summary and feature list were written by an AI model that read the project's README and public documentation pages. Each feature links to the documentation it came from; stars, license and language come straight from the GitHub API. The model does not read the source code, and the analysis is refreshed when the project is re-analysed. Learn more on our About page.
