CROMA Remote Sensing Representations with Contrastive Radar-Optical Masked Autoencoders

• 论文名称: CROMA: Remote Sensing Representations with Contrastive Radar-Optical Masked Autoencoders
• Link: https://arxiv.org/pdf/2311.00566
• Published in: NeurIPS 2023
• Type: Pretrain
• Code/Project: https://github.com/antofuller/CROMA
• 备注: 对比学习+重建，多光谱+孔径雷达
• Backbone: ViT
• Backbone 1: ViT-B/ViT-L
• 下游任务: Classification, Semantic Segmentation
• 下游任务 1: Classification,Segmentation
• Short Summary: 分别编码被遮盖的多光谱光学和合成孔径雷达样本，这些样本在空间和时间上是对齐的，并执行跨模态对比学习。另一个编码器融合这些传感器，生成联合多模态编码，然后通过一个轻量级解码器预测被遮盖的补丁。