This dataset contains a labeled set of High Resolution Imaging Science Experiment (HiRISE) image tiles that either do or do not contain visible indications of frost presence. The dataset was created using Labelbox to draw polygonal annotations over subframes extracted from HiRISE observations. Then, each subframe was broken into 299x299 pixel tiles at 0.5 m/pixel resolution, and a majority vote across annotations is used to determine whether each tile is given a "frost," "background," or "ambiguous" label. Ambiguous labels are assigned if there is not a majority agreement about the label of the tile given the polygons. All tiles from "background" subframes (taken during summer months when no frost is present) are assigned a "background" label. Corresponding to each tile is a JSON label file containing metadata from overlapping annotations. There is also a TensorFlow TFRecord object saved for each subframe that contains the set of all tiles extracted from that subrame.

{"references": ["Alfred McEwen. 2007. Mars Reconnaissance Orbiter High Resolution Imaging Science Experiment, Reduced Data Record, MRO-M-HIRISE-3-RDR-V1.1. NASA Planetary Data System."]}