Review on self-supervised image recognition using deep neural networks
作者：
Highlights：
• Self-supervised learning and other learning methods.
• Self-supervised and supervised learning pipeline.
• Description of various handcrafted pretext tasks for self-supervised learning.
• Self-supervised methods that follow contrastive or instance discrimination approach.
• Performance comparisons of the discussed self-supervised techniques on evaluation tasks such as image classification and object detection.
• Practical considerations and open challenges while performing self-supervised learning in image recognition tasks.
摘要
•Self-supervised learning and other learning methods.•Self-supervised and supervised learning pipeline.•Description of various handcrafted pretext tasks for self-supervised learning.•Self-supervised methods that follow contrastive or instance discrimination approach.•Performance comparisons of the discussed self-supervised techniques on evaluation tasks such as image classification and object detection.•Practical considerations and open challenges while performing self-supervised learning in image recognition tasks.
论文关键词：Self-supervised learning,Unsupervised learning,Semi-supervised learning,Transfer learning,Deep learning,Pretext tasks,Convolutional neural network,Contrastive learning,Online clustering
论文评审过程：Received 19 October 2020, Revised 14 April 2021, Accepted 26 April 2021, Available online 29 April 2021, Version of Record 5 May 2021.
论文官网地址：https://doi.org/10.1016/j.knosys.2021.107090
