Input and output data (images + boulder labels, model setup, model weights and more) for t

Planetary images and labeled boulders from Earth, Mars, and the Moon, supporting the manuscript "Automatic characterization of boulders on planetary surfaces from high-resolution satellite images". The dataset includes 8.8 GB of raw rasters and shapefiles, plus pre-processed training data and model weights for a Mask R-CNN model named BoulderNet. It was authored by Nils C. Prieur of the University of Oslo.

Use Cases

• Training object detection models for boulder identification based on the provided labeled shapefiles and satellite images.
• Benchmarking geospatial computer vision algorithms using the pre-processed train, validation, and test image patches.
• Transfer learning for planetary surface analysis using the provided Mask R-CNN model weights and configuration files.

Strengths

• Includes data from three distinct planetary bodies: Earth, Mars, and the Moon.
• Provides a complete model training pipeline with 624.7 MB of model weights, configuration, and logs.
• Contains 950 pre-processed image patches for machine learning, as indicated in the description.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

University of Oslo, associated with the manuscript by Nils C. Prieur.

Collection Method

Manually digitized boulder outlines from high-resolution satellite images.

Geography

Earth, Mars, Moon