S5Mars: A High-Quality Dataset and Semi-Supervised Learning Framework for Efficient Mars Terrain Semantic Segmentation

The core message of this article is to present a new high-quality dataset, S5Mars, and a semi-supervised learning framework tailored for efficient Mars terrain semantic segmentation. The proposed method addresses the challenges of limited annotated data and the ineffectiveness of existing augmentations for Mars images, achieving state-of-the-art performance.