imgaug and Augmentor
These are competitors offering overlapping functionality for image augmentation pipelines—both provide transformation libraries for training data, though imgaug dominates in adoption and API flexibility while Augmentor emphasizes simplicity and GUI-based workflows.
About imgaug
aleju/imgaug
Image augmentation for machine learning experiments.
Supports augmentation of diverse annotation types—heatmaps, segmentation maps, keypoints, bounding boxes, and polygons—with automatic coordinate alignment so transformations apply consistently across all data types. Provides 50+ geometric and photometric operations (affine, perspective, contrast, noise, blur) optimized for batch processing, with composable pipelines that apply random augmentations in configurable orders. Integrates seamlessly with numpy arrays and major ML frameworks through a NumPy-based architecture, enabling deterministic augmentation through seed control.
About Augmentor
mdbloice/Augmentor
Image augmentation library in Python for machine learning.
Provides framework-agnostic pipeline-based augmentation with stochastic probability controls and multi-threading support for efficient batch generation. Supports synchronized augmentation of ground truth masks and segmentation labels through dedicated `Pipeline` and `DataPipeline` classes. Includes generators for Keras and PyTorch integration, enabling on-the-fly augmentation during model training without disk I/O.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work