IWSeg-SAR-Poison

该数据集专为从合成孔径雷达（SAR）图像中分割内陆水体而设计，包含了用于增强训练鲁棒性的对抗性样本。数据集不仅包含了未损坏的图像，还包含了受到对抗性攻击损坏的图像，这为评估模型的鲁棒性提供了可能。该数据集的任务是对SAR图像中的内陆水体进行分割。This dataset is specifically tailored for inland water body segmentation from Synthetic Aperture Radar (SAR) images, and incorporates adversarial samples to enhance the robustness of model training. It includes both uncorrupted images and images corrupted via adversarial attacks, which facilitates the assessment of model robustness. The defined task of this dataset is to segment inland water bodies from SAR images.

搜集汇总

背景与挑战

背景概述

IWSeg-SAR-Poison是一个用于内陆水体分割的合成孔径雷达（SAR）图像数据集，专注于研究形态学标签噪声对深度学习模型鲁棒性的影响。数据集包含清洁的掩码和通过形态学操作模拟人工标注错误的对抗性损坏掩码，旨在评估和比较不同分割模型在标注噪声下的性能。