Change-Aware Sampling and Contrastive Learning for Satellite Images

• 论文名称: Change-Aware Sampling and Contrastive Learning for Satellite Images
• Link: https://openaccess.thecvf.com/content/CVPR2023/html/Mall_Change-Aware_Sampling_and_Contrastive_Learning_for_Satellite_Images_CVPR_2023_paper.html
• Published in: CVPR 2023
• Type: Pretrain
• Code/Project: https://github.com/utkarshmall13/CACo
• 备注: 自监督，对比学习
• Backbone: Resnet
• Backbone 1: CNN（ResNet-18, ResNet-50）
• 下游任务: Classification, Change Detection, Semantic Segmentation
• 下游任务 1: Classification（Landcover classification），Change detection，Semantic segmentation，Change Events
• Short Summary: 季节变化损失与SeCo相同，使用同一地点几个月时间内的图片作为正样本，其他地点的图片作为负样本。长时间变化用几个月内的做正样本，长时间后的图片做负样本。