ESSA PyTorch Model Checkpoint and Shapefiles Containing Pit and Skylight Detections on the Moon

This dataset contains the PyTorch model checkpoint file (also called the state_dict) of ESSA - a Mask R-CNN instance segmentation Deep Learning model trained to detect pits and skylights within Lunar Reconnaissance Orbiter Narrow Angle Camera (LROC NAC) imagery of the Moon. Instructions on how to load a model checkpoint in order to re-train or infer it on your own data can be found here, or visit ESSA's GitHub page. The .zip file ESSA_shapefiles.zip contains the locations of the pit and skylight detections made by ESSA within eight Regions of Interest for searching for potential cave entrances ('ESSA_detections.shp'). These detections are in ESRI shapefile format ready for viewing in GIS software such as QGIS or ArcGIS. We also provide shapefiles of the mapping of Lunar rilles within the Marius Hills region of the Moon ('marius_hills_rilles.shp'). Both of these sets of shapefiles have attribute fields containing the confidence scores that they were detected with and the images that they were found to be within. Furthermore, the .csv file training_images.csv (as the name suggests) provides a full list of the LROC NAC and MRO HiRISE image products that were the source of the tiles seen by ESSA during the training process. The file also provides the source of the 'ground truth' for these images, which are as follows. Moon: mare, impact_melt and highlands - Lunar Pit Atlas (LPA; Wagner & Robinson, 2021) apollo - Mosaics of Apollo landing sites (Klem et al., 2014 and Haase et al., 2019) rilles - Lunar lava-tube-related rilles (Hurwitz et al., 2013) Mars: APC, pit, sky, srp - Mars Global Cave Candidate Catalog (MGC3; Cushing, 2015) pce - Potential Cave Entrances (Watson and Baldini, 2024)