Mars-Bench: mb-domars16k

This is a multi-class classification dataset designed for geomorphologic feature recognition on Mars using imagery from the CTX sensor. It consists of 15 classes grouped into five thematic categories: (1) Aeolian Bedforms: Aeolian Curved, Aeolian Straight; (2) Topographic Landforms: Channel, Cliff, Mounds, Ridge; (3) Slope Features: Gullies, Mass Wasting, Slope Streaks; (4) Impact Landforms: Crater, Crater Field; and (5) Basic Terrain: Mixed Terrain, Rough Terrain, Smooth Terrain, Textured Terrain. This is one of the largest and diverse orbital datasets in terms of a number of classes. Hence, the dataset presents a unique challenge due to its class granularity, significant variability within classes, and subtle differences between classes, making it valuable for evaluating models on fine-grained classification and generalization. To explore all Mars-Bench resources, visit our project website: https://mars-bench.github.io/