Post

Feature Guided Masked Autoencoder for Self-supervised Learning in Remote Sensing

  • 论文名称: Feature Guided Masked Autoencoder for Self-supervised Learning in Remote Sensing
  • Link: https://arxiv.org/abs/2310.18653
  • Published in: Arxiv 2023
  • Type: Pretrain
  • Code/Project: https://github.com/zhu-xlab/FGMAE
  • 备注: 多模态,掩码
  • Backbone: ViT
  • Backbone 1: ViT
  • 下游任务: Classification
  • 下游任务 1: classification
  • Short Summary: MAE往往过度关注像素细节,从而限制了模型的语义理解能力,特别是对于噪声较大的SAR图像。提出了特征引导掩码自动编码器(FG-MAE):重建多光谱图像的方向梯度直方图(HOG)和归一化差异指数(NDI)的组合,以及重建SAR图像的HOG。
This post is licensed under CC BY 4.0 by the author.