Mars Orbital Image (HiRISE) Labeled Data Set: 64,947 Landmark Images

64,947 labeled images of Martian surface landmarks were extracted and processed from 232 HiRISE orbital browse images. The dataset was created by Gary Doran of the Jet Propulsion Laboratory, with images cropped to include landmarks with a margin and resized to 227x227 pixels. A subset of 9,022 original landmarks was augmented via rotation, flipping, and brightness adjustments to generate the full collection.

Use Cases

• Training image classification models to identify Martian geological features like dunes and craters based on the described landmark classes.
• Developing object detection algorithms for locating specific surface phenomena such as slope streaks or impact ejecta in orbital imagery.
• Benchmarking data augmentation techniques using the described methods of rotation, flipping, and brightness adjustment.
• Studying the visual characteristics of polar phenomena like spiders and Swiss cheese mentioned in the class discussion.

Strengths

• Contains 64,947 processed and labeled images, providing substantial volume for model training.
• Includes a detailed class dictionary mapping IDs to semantic names for features like bright dune, dark dune, crater, and slope streak.
• Provides pre-defined train/validation/test splits and upsampling information in a separate label file.
• Images are consistently pre-processed to 227x227 pixels with a defined cropping margin.

Limitations

• Column-level documentation is absent; field semantics for label files must be inferred after download.
• Last update date is unknown; freshness unverified.
• The dataset's spatial and temporal coverage is not specified, which may limit understanding of geographic or seasonal bias.

Provenance

Source

Jet Propulsion Laboratory

Collection Method

Landmarks were detected and extracted from HiRISE (High Resolution Imaging Science Experiment) browse images, then cropped, resized, and augmented.

Geography

Mars