Lunar Technosignatures

The "Lunar Technosignatures" dataset is derived from high-resolution images captured by the Lunar Reconnaissance Orbiter (LRO), a spacecraft launched in 2009. The dataset focuses on images taken by the Narrow Angle Camera (NAC), which provides a resolution of approximately 0.5 meters per pixel. The NAC images are processed into 224x224 pixel patches with a stride of 28, excluding a 408-pixel radius around the descent stage of lunar landing sites. This preprocessing yields 492,070 training images for Apollo 15 and 518,200 training images for Apollo 17. Additionally, test images are generated from the excluded landing sites using a stride of 8, resulting in approximately 5,476 test images per site. The dataset is ideal for high-resolution lunar surface analysis and technosignature detection.

Natural Oak

The "NaturalOak" dataset originates from a production line for natural wood flooring. It consists of high-resolution images from 36 wooden planks, captured in a laboratory using RGB and reflex camera technology. The reflex channel technique highlights surface defects such as cracks, bumps, and holes by casting light at a sharp angle and recording it from the opposite side. The dataset focuses on surface structure defects like chippings, cracks, and other imperfections, which are more visible in the reflex channel compared to RGB images.

The dataset was split into sub-images of size 1024x1024 pixels, resulting in 247 training points without defects and 85 test points, 61 of which contain defects. This makes the dataset comparable in size to the wood category of the MVTecAD dataset. Each defect has been manually annotated, providing ground truth for accurate evaluation.