Self-supervised visual learning in the low-data regime: a comparative evaluation

arXiv:2404.17202v1 Announce Type: new
Abstract: Self-Supervised Learning (SSL) is a valuable and robust training methodology for contemporary Deep Neural Networks (DNNs), enabling unsupervised pretraining on a `pretext task' that does not require ground-truth labels/annotation. This allows efficient representation learning from massive amounts of unlabeled training data, which in turn leads to increased accuracy in a `downstream task' by exploiting supervised transfer learning. Despite the relatively straightforward conceptualization and applicability of SSL, it is not always feasible to collect and/or to …

abstract annotation arxiv cs.cv data enabling evaluation ground-truth labels leads low massive methodology networks neural networks pretraining representation representation learning robust self-supervised learning ssl supervised learning training training data truth type unsupervised visual