Cross-Scale MAE A Tale of Multiscale Exploitation in Remote Sensing

• 论文名称: Cross-Scale MAE: A Tale of Multiscale Exploitation in Remote Sensing
• Link: https://openreview.net/pdf?id=5oEVdOd6TV
• Published in: NeurIPS 2023
• Type: Pretrain
• Code/Project: https://github.com/aicip/Cross-Scale-MAE
• 备注: MAE+对比学习
• Backbone: ViT
• Backbone 1: ViT-L
• 下游任务: Classification, Semantic Segmentation
• 下游任务 1: classification and segmentation
• Short Summary: 在预训练过程中，Cross-Scale MAE采用了尺度增强技术，并通过对比损失和生成损失实施跨尺度一致性约束，以确保一致且有意义的表示，适用于广泛的下游任务。