RandAugment for Image Classification for Improved Robustness on the 🤗Hub!
Excerpt from the Tutorial:
Data augmentation is a very useful technique that can help to improve the translational invariance of convolutional neural networks (CNN). RandAugment is a stochastic vision data augmentation routine composed of strong augmentation transforms like color jitters, Gaussian blurs, saturations, etc. along with more traditional augmentation transforms such as random crops.
Recently, it has been a key component of works like Noisy Student Training and Unsupervised Data Augmentation for Consistency Training. It has been also central to the success of EfficientNets.
About The dataset
The model was trained on CIFAR-10, consisting of 60000 32x32 color images in 10 classes, with 6000 images per class. There are 50000 training images and 10000 test images.