Improving Self-supervised Learning with Automated Unsupervised Outlier Arbitration
Augmenting images training data to add variability doesn't always end well since the augmentation changes the semantic content of the image. Here's how to solve it.