Self-Supervised Contrastive Representation Learning in Computer Vision

Although its origins date a few decades back, contrastive learning has recently gained popularity due to achievements in self-supervised learning, especially computer vision. Supervised usually requires decent amount of labeled data, which is not easy obtain for many applications. With we can use inexpensive unlabeled data and achieve training on pretext task. Such helps us learn powerful representations. In most cases, downstream task, fine-tuned with the available data. this study, review common tasks vision present latest techniques, are implemented as Siamese neural networks. Lastly, case study where was applied representations semantic masks images. Performance evaluated an image retrieval task results reveal that, accordance findings literature, fine-tuning showed best performance.