BROWSE
Detail
Conference Paper
컴퓨터비전에서의 최신 자가지도학습기법 동향
Cited - time in 
Share 
Share
- Issue Date
- 2023-06
- Citation
- 대한전자공학회 학술 대회 (하계) 2023, pp.1-5
- Publisher
- 대한전자공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- Recently, self-supervised learning (SSL) without supervision (i.e., human annotation) in the computer vision domain has drawn huge popularity due to its superior performance than deep supervised learning on downstream tasks such as object detection and segmentation. Thus, it is more likely to use the pretrained models by self-supervised learning than supervised learning. Specifically, as masked language modeling (MLM) in natural language processing (NLP) shows powerful representation in the pre-training phase, masked image modeling (MIM) has emerged in SSL for the computer vision domain. From this context, we try to review recent self-supervised learning algorithms released since 2020 in terms of pre-text tasks, architecture, and backbone network and provide several discussion points.
- KSP Keywords
- Backbone Network, Computer Vision(CV), Human annotation, Image modeling, Language modeling, Natural Language Processing(NLP), Pre-Training, Supervised learning algorithm, object detection, self-supervised learning, superior performance