The 8-band images on their own have a resolution per pixel of ~1.3m for WV-3 images and ~2m for WV-2 images. The 8-band, multispectral images include the following bands: Coastal Blue, Blue, Green, Yellow, Red, Red Edge, Near Infrared 1 (NIR1), and Near Infrared 2 (NIR2).
SpaceNet provides panchromatic images which can map to gray-scale images at a resolution per pixel of 0.3m for WV-3 and 0.5m for WV-2. It should be noted that downloading from AWS will require an AWS account which, in turn, requires credit-card information. However, the dataset is readily available for use within the MLRG shared dataset folder: /mnt/data/datasets/SpaceNet. Download instructions can be found here (SpaceNet on AWS). The dataset is publicly available through Amazon Web Services (AWS) for free. A high resolution example from the WV-3 is shown here: Images from Rio de Janeiro were taken with the WV-2 satellite, whereas the remaining cities’ images were taken using the higher resolution WV-3. Images come from five cities or “Areas of Interest” (AOI), Rio de Janeiro (AOI_1), Las Vegas (AOI_2), Paris (AOI_3), Shanghai (AOI_4) and Khartoum (AOI_5). An example of an image and its building footprint ground-truth can be seen below:
#Nvidia spacenet Patch#
Building footprints for each patch (corresponding to a 200m x 200m image) are stored as similarly named GeoJSON files which provide global coordinates that define closed polygons indicating the location and extent of a building within an image. For each patch, the dataset includes panchromatic, RGB and 8-band multispectral images. The data is composed of multispectral images of cities (200m x 200m patches) with corresponding ground-truth building location data. The SpaceNet dataset is a body of 17355 images collected from DigitalGlobe’s WorldView-2 (WV-2) and WorldView-3 (WV-3) multispectral imaging satellites and has been released as a collaboration of DigialGlobe, CosmiQ Works and NVIDIA. Images from on High - The SpaceNet Dataset Dataset Overview
0 kommentar(er)
