A 1m x 1m DEM of the greater campus area. Created by MIL staff from 11 separate files created by NOAA as part of a larger coastal LiDAR survey. Original pixel values are preserved (and are in meters). Stretched values are also included in the attribute table.

Source data: ftp://coast.noaa.gov/pub/DigitalCoast/raster2/elevation/California_Lidar_DEM_2009_1131 Source metadata for larger dataset: http://coast.noaa.gov/dataservices/Metadata/TransformMetadata?u=http://coast.noaa.gov/data/Documents/Metadata/Imagery/harvest/ca2010_coastal_dem.xml&f=html

The datasets were generated for the Upper Choptank River Watershed (UCRW) in Maryland and Delaware. Depressions and the total stream network were derived from a lidar DEM collected during April to June 2003 (vertical accuracy root mean square error (RMSE) = 14.3 cm) and March to April 2006 (vertical accuracy RMSE = 18.5 cm) for Maryland (1 m resolution) and April 2007 (vertical accuracy RMSE = 18.5 cm) for Delaware (3 m resolution)). Depressions were identified using the Stochastic Depression Analysis Tool in Whitebox GAT. A random sample from the potential error values was iteratively added to the original DEM prior to depressions being filled and identified. Cells were considered part of a depression if identified as such in 80% of the 20 iterations. An edge preserving smoothing filter with a 3-pixel by 3-pixel window was applied to reduce noise, or random error, within the final depression raster. Depressions smaller than 50 m2 were removed. To map the stream network, flow accumulation was calculated on the filled lidar DEM using the FD8 flow accumulation algorithm. Flow accumulation was thresholded at 50,000 m2, a decision point guided by field-based points. Three Landsat images were used to create a surface-water extent map from Landsat including Landsat-8 images collected on April 4, 2015 (p14r33) and April 11, 2015 (p15r33) and a Landsat-7 ETM+ image collected on April 12, 2015 (p14r33). Surface-water was identified using the Matched Filtering algorithm. The output values were linearly stretched and a Frost filter with a 3-pixel by 3-pixel window was applied. Landsat pixels with a per-pixel fraction of >0.4 were classified as inundated.

The "Zurich Summer v1.0" dataset is a collection of 20 chips (crops), taken from a QuickBird acquisition of the city of Zurich (Switzerland) in August 2002. QuickBird images are composed by 4 channels (NIR-R-G-B) and were pansharpened to the PAN resolution of about 0.62 cm GSD. We manually annotated 8 different urban and periurban classes : Roads, Buildings, Trees, Grass, Bare Soil, Water, Railways and Swimming pools. The cumulative number of class samples is highly unbalanced, to reflect real world situations. Note that annotations are not perfect, are not ultradense (not every pixel is annotated) and there might be some errors as well. We performed annotations by jointly selecting superpixels (SLIC) and drawing (freehand) over regions which we could confidently assign an object class.

The dataset is composed by 20 image - ground truth pairs, in geotiff format. Images are distributed in raw DN values. We provide a rough and dirty MATLAB script (preprocess.m) to:

i) extract basic statistics from images (min, max, mean and average std) which should be used to globally normalize the data (note that class distribution of the chips is highly uneven, so single-frame normalization would shift distribution of classes).

ii) Visualize raw DN images (with unsaturated values) and a corresponding stretched version (good for illustration purposes). It also saves a raw and adjusted image version in MATLAB format (.mat) in a local subfolder.

iii) Convert RGB annotations to index mask (CLASS \in {1,...,C}) (via rgb2label.m provided).

iv) Convert index mask to georeferenced RGB annotations (via rgb2label.m provided). Useful if you want to see the final maps of the tiles in some GIS software (coordinate system copied from original geotiffs).

Some requests from you

We encourage researchers to report the ID of images used for training / validation / test (e.g. train: zh1 to zh7, validation zh8 to zh12 and test zh13 to zh20). The purpose of distributing datasets is to encourage reproducibility of experiments.

Acknowledgements

We release this data after a kind agreement obtained with DigitalGlobe, co. This data can be redistributed freely, provided that this document and corresponding license are part of the distribution. Ideally, since the dataset could be updated over the time, I suggest to distribute the dataset by the official link from which this archive has been downloaded.

We would like to thank (a lot) Nathan Longbotham @ DigitalGlobe and the whole DG team for his / their help for granting the distribution of the dataset.

We release this dataset hoping that will help researchers working in semantic classification / segmentation of remote sensing data in comparing to other state-of-the-art methods using this dataset as well in testing models on a larger and more complete set of images (with respect to most benchmarks available in our community). As you can imagine, it has been a tedious work in preparing everything. Just for you.

If you are using the data please cite the following work

• Volpi, M. & Ferrari, V.; Semantic segmentation of urban scenes by learning local class interactions, In IEEE CVPR 2015 Workshop "Looking from above: when Earth observation meets vision" (EARTHVISION), Boston, USA, 2015.

Introduction

The aurora images in this dataset are from the all-sky imagers at the Yellow River Station (YRS). And it is used in the paper Automatically sketching Auroral Skeleton Structure in All-sky Image for Measuring Aurora Arcs.

This dataset consists of the following sections:

• N20040116G.zip: 4,319 observation images as of January 16, 2004;
• Raw data and reading code.zip: 8634 raw data items of December 23, 2003, and reading code;
• 80 imgs&annotations.zip: 80 ASI images and the corresponding manual annotations with pixel-level masks.

In the following sections, the data source, acquisition instruments, data characteristics, preprocessing methods and naming conventions of this dataset will be introduced.

Data source:

The auroral images used in this study are ground-based observation data acquired by three-wavelength (427.8, 557.7, and 630.0 nm) all-sky imagers at YRS in Ny-Ålesund, Svalbard. The geographic coordinates of YRS are 78.92°N, 11.93°E, and the corrected geomagnetic latitude is 76.24°, MLT≈UT+3 h.

Data acquisition instruments:

Data were obtained from all-sky CCD auroral imagers. The system can record 24-hour auroral activities from October to March of the following winter each year.

Data characteristics:

The time resolution of the observation results of two consecutive frames is 10 s. The camera uses a 512×512 square pixel array, and the geographic zenith is located near the center of the field of view (FOV). The coverage diameter of the all-sky FOV is approximately 1000 km. The spatial resolution is from 1.1 km at the zenith to 36 km at the horizon at an altitude of 150 km. All aurora images are provided in .bmp image format.

Data preprocess:

• Subtracting dark current and rescaling. Dark current is deemed as system noise caused by equipment. It is removed from images before further operation. Every image is then stretched with a cutoff value of 4000 and rescaled from 16 to 8 bits.
• Masking and cropping. A circle mask with a radius of 220 pixels is applied to cut off the outer regions where significant wide-angle distortion happens and may contain YRS lights. Then, the original image size of 512×512 is cropped to 440×440 pixels.
• Rotation. We rotated the images counter-clockwise by 61.11 with the north being upward.

Data naming convention:

Each auroral data has a unique file name that contains the N/S, date, band, and time information.

Filename example: N20031221G081212.

• N: North pole
• 20031221: December 21, 2003
• G: G-band
• 081212: 8:12:12 UT