Beacon-aug¶

Why Beacon_aug¶

Incorporate the largest number of image augmentation operations(> 300) from 8 popular libraries
Seamless cross library exchanging over different libraries
Adobe Featured customized functions of both parametric and GAN based transformations designed by Adobe Researchers
Run external AI function inferencing as easy as general augmentation
Advanced transformation pipelines for complex tasks (e.g. segmentation, detection, GAN training, network robustness)
Support various input formats : np.array,PIL , Torch.tensor
Extend the high-level attributes from Albumentations to other libraries by dynamic loading
Differentiable check, anti-aliasing for operators
Fast-visualization of the augmentation pipeline
Easy to add customized functions for public contributors

Library	Beacon_Aug	Imgaug	Albumentations	torchvision	Keras	Augly	mmcv	ImageNet-C	kornia
Image	✓	✓	✓	✓	✓	✓	✓	✓	✓
Mask	✓	✓	✓	✓	✓	✓	✓	x	✓
Bounding Box	✓	✓	✓	x	x	x	✓	x	x
Keypoints	✓	✓	✓	x	x	x	✓	x	x
Paired Transformation	✓	✓	✓	x	x	x	✓	x	x
Parameter Outputs	✓	x	✓	x	x	x	✓	x	x
Differentiable	✓	x	x	✓	x	x	x	x	✓
Customized Function	✓	x	✓	x	x	x	x	x	x
GAN-based Function	✓	x	x	x	x	x	x	x	x
AutoAugment	✓	x	x	x	x	x	x	x	x
RandAug	✓	x	x	x	x	x	x	x	x
Num of Transformations	>378	107	70	37	11	31		19
Cross Library Supports	Imgaug,Torch, Keras,Augly, Imagenet-c,mmcv, Albumentations	CV2, PIL, Skimage	CV2, Imgaug Torch	PIL	PIL	PIL	PIL	PIL	Torch
Average Run Time	Depend on libraries	13.6 ms	2.1 ms	5.2 ms	59.3 ms	25.3 ms

Main module building: Rebecca Li, Yannick Hold-Geoffroy, Geoffrey Oxholm
Customized functions and advanced properties contributing: Richard Zhang, Maksym Andriushchenko, Krishna Kumar Singh, Zhifei Zhang