Mars-Bench: mb-surface_cls

This is a multi-class classification dataset consisting of surface imagery captured by the Mastcam and MAHLI instruments aboard the Curiosity rover. It comprises 36 classes, making it the largest and most diverse surface imagery dataset included in Mars-Bench. The classes span a wide range of surface elements, including rover components, scientific instruments, geologic features, and environmental elements. The full list includes: Alpha Particle X-Ray Spectrometer (APXS), APXS Calibration Target, Arm Cover, Artifact, ChemCam Calibration Target, CheMin Inlet Open, Close-Up Rock, Distant Landscape, Drill, Drill Holes, Dust Removal Tool (DRT), DRT Spot, Float Rock, Ground, Horizon, Inlet, Layered Rock, Light-Toned Veins, MAHLI, MAHLI Calibration Target, Mastcam, Mastcam Calibration Target, Night Sky, Observation Tray, Portion Box, Portion Tube, Portion Tube Opening, REMS-UV, Rover Rear Deck, Sand, Scoop, Sun, Turret, Wheel, Wheel Joint, Wheel Tracks. Images were labeled using the IDAR tool, incorporating annotations from both domain experts and volunteers. For ambiguous samples, class prioritization rules were applied to assign the most representative label. As with many real-world planetary datasets, mb-surface_cls is highly imbalanced, with a large proportion of samples belonging to the Ground class. To explore all Mars-Bench resources, visit our project website: https://mars-bench.github.io/