Mars-Bench: mb-landmark_cls

This is a multi-class classification dataset derived from orbital HiRISE imagery. It classifies into 8 surface feature classes: Bright Dune, Crater, Dark Dune, Impact Ejecta, Slope Streak, Spider, Swiss Cheese, and Other. The class distribution is highly imbalanced, with “Other” comprising the majority of samples and Impact Ejecta being the minority class. Landmarks were extracted using a dynamic salience-based method. Labels were generated via a mix of volunteer crowdsourcing and expert validation, with additional calibration techniques applied to improve reliability. To explore all Mars-Bench resources, visit our project website: https://mars-bench.github.io/本数据集为基于轨道高分辨率成像科学实验（HiRISE）影像构建的多分类任务数据集，共包含8类地表特征类别：亮沙丘（Bright Dune）、撞击坑（Crater）、暗沙丘（Dark Dune）、撞击溅射物（Impact Ejecta）、坡痕（Slope Streak）、蜘蛛状地貌（Spider）、瑞士奶酪地貌（Swiss Cheese）以及其他类（Other）。该数据集的类别分布高度不平衡，其中"Other"类样本占绝大多数，撞击溅射物为样本量最少的少数类别。本数据集的地标特征采用基于动态显著性的方法提取，标签通过志愿者众包与专家验证相结合的方式生成，并辅以额外的校准技术以提升标注可靠性。 如需探索全部Mars-Bench相关资源，请访问项目官网：https://mars-bench.github.io/