imgaug
Python library that creates many slightly altered copies of training images for machine learning.
- Stars
- 14.7k
- Forks
- 2.4k
- Licence
- MIT
- Language
- Python
- Last update
- 2.3 years ago
- Type
- Library
No commits in over a year — check for forks or alternatives before adopting.
What it is
imgaug is a Python library that takes a set of input images and produces a larger set of slightly altered versions, such as rotated, cropped, noisy, or recolored copies. It is used to expand training data for machine learning projects. It can also transform matching annotations like segmentation maps, keypoints, and bounding boxes so they stay aligned with the changed images.
- Apply random rotation, scaling, and flips to training images along with their bounding boxes
- Add Gaussian noise, contrast changes, or blur to make a model more robust
- Crop or pad images while keeping segmentation maps aligned
- Run augmentations across multiple CPU cores for faster batch processing
Great for
- Machine learning engineers training image models on a limited dataset
- Researchers who need augmentations that keep labels such as bounding boxes or keypoints aligned
- Developers building Python image pipelines who want many augmentation types in one library
Not ideal for
- People who only need a simple flip or rotation, since OpenCV or torchvision may be lighter for that
- Projects needing non-Python environments, since the README describes a Python library
- Users on Python 2.7 or very new Python versions without checking compatibility, since the README only states support for 2.7 and 3.4+
How to get it
- 1
Install with a package manager
You'll need: Python (README mentions 2.7 and 3.4+; check current compatibility)
Clone the repository
git clone https://github.com/aleju/imgaug - 2
Run it
- Install the library following the Installation section of the README
- Import imgaug and create an augmentation pipeline with the augmenters you need
- Pass your images and annotations through the pipeline to get augmented outputs
- 3
Get the most out of it
- Use the probability distributions as parameters to control how strong each augmentation is
- Use the helper functions to draw heatmaps, segmentation maps, or keypoints for checking results
- Use multi-core augmentation to speed up large batches
- Read the documentation at imgaug.readthedocs.io for dtype support and examples
Downloads
Strengths
- Covers many augmentation types, including affine, perspective, noise, and cropping/padding
- Keeps heatmaps, segmentation maps, keypoints, bounding boxes, and polygons aligned with images
- Released under the MIT license with a large community following (about 14.7k stars)
Trade-offs
- The latest push is from 2024-07-30 and the README shows Travis CI badges, so maintenance activity should be checked before relying on it
- The README mentions Python 2.7 and 3.4+ support, which may not reflect current Python versions
- The README is truncated here, so detailed install steps and API usage were not visible
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| imgaug (this) | 14.7k | Library | intermediate | 2.3 years ago | Machine learning engineers training image models on a limited dataset |
| albumentations | 15.3k | Library | beginner | 1.4 years ago | Machine learning engineers training image classification or segmentation models with limited data |
| wandb | 11.3k | Library | beginner | 4 days ago | Machine learning engineers who want to compare training runs over time |
| CLIP | 34.6k | AI model / agent | intermediate | 8 months ago | Python developers experimenting with image classification using text labels |
| shap | 25.8k | Library | intermediate | 5 days ago | Data scientists who need to explain tree-based models such as XGBoost, LightGBM, or CatBoost |
| deepface | 23.6k | Library | intermediate | 6 days ago | Python developers prototyping face verification or face search |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.