DYNO

From DINO to DYNO

Training Contrastive Learning models with image augmentations is critical to allow for downstream image classication to be augmentation-invariant. Dynamic augmentation scheduling for Contrastive Learning has not been researched significantly. In From DINO to DYNO, the use of a dynamic augmentation scheduler is investigated to determine whether classification results can increase on the DINOv2 model. This research encompasses comparisons to a previously tried method from Peking University and static augmentation.