SimMars6K

Due to the lack of actual Mars rover image data and the high cost of manually labeling training samples, we created a simulated Mars rover image dataset, SimMars6K, containing 6,325 pairs of images covering different terrain types with pixel-level annotations. This dataset was created using the OAISYS simulator developed by the German Aerospace Center, based on the open source software Blender with its rendering engine and programmable Python API. Referring to the Navigation and Terrain Camera (NaTeCam) of China's Zhurong Mars rover, the stereo camera is set 1.2 meters above the surface with a baseline of 270 mm, a field of view (FOV) of 46.5°×46.5°, and an image resolution of 512×512. In addition, this dataset also contains corresponding depth images, two-level semantic annotations, and instance segmentation masks that can be valuable for studying various algorithms, including those for the Mars rover's terrain perception, navigation and localization, and path planning. To use this dataset, please cite: C. Ma, Y. Li, J. Lv, Z. Xiao, W. Zhang and L. Mo, "Automated Rock Detection from Mars Rover Image Via Y-Shape Dual-Task Network with Depth-Aware Spatial Attention Mechanism," in IEEE Transactions on Geoscience and Remote Sensing, doi: 10.1109/TGRS.2024.3371684.