PRIME: A few primitives can boost robustness to Common Corruptions

This upload contains the neural networks used in the paper "PRIME: A Few Primitives Can Boost Robustness to Common Corruptions". PRIME is a generic, plug-n-play data augmentation scheme that consists of simple families of max-entropy image transformations for conferring robustness to common corruptions. PRIME leads to significant improvements in corruption robustness on multiple benchmarks.

The networks are already pre-trained on the CIFAR-10, CIFAR-100, ImageNet-100 and ImageNet using PRIME. For ImageNet/100, we also provide a model that has been trained by combining DeepAugment + PRIME. The networks are implemented in PyTorch. The ImageNet/100 models can be loaded directly using the model implementations provided by Torchvision. For loading the CIFAR-10/100 models, please use the model definitions provided here.

More information regarding the implementation and usage of PRIME is provided in the official GitHub repository.