Voxel51/S5Mars

S5Mars（火星语义分割半监督学习）是一个高分辨率的火星地形语义分割数据集，包含6,000张由火星科学实验室（MSL）好奇号探测器拍摄的图像。该数据集具有稀疏但高置信度的注释，旨在支持火星地形分析的半监督学习方法。数据集解决了火星图像高质量注释有限的问题，提供了精心策划的基于置信度的稀疏注释，使得研究可以利用标记和未标记的火星地形数据进行自主探测器导航和路径规划的研究。每张图像都标注了像素级的语义分割掩码，涵盖了火星地形中常见的9种地形类别。

S5Mars (Semi-SuperviSed learning on Mars Semantic Segmentation) is a high-resolution Mars terrain semantic segmentation dataset containing 6,000 images captured by the Mars Science Laboratory (MSL) Curiosity rover. The dataset features sparse, high-confidence annotations designed to support semi-supervised learning approaches for Martian terrain analysis. The dataset addresses the challenge of limited high-quality annotations for Mars imagery by providing carefully curated, confidence-based sparse annotations. This enables research into semi-supervised learning methods that can leverage both labeled and unlabeled Mars terrain data for autonomous rover navigation and planning. Each image is annotated with pixel-level semantic segmentation masks covering 9 terrain classes commonly encountered in Martian landscapes.