isp-uv-es/mdas-taco

MDAS (Augsburg multimodal data set)

This is a repackaging, not a new dataset. It is MDAS (Augsburg multimodal data set) by Technical University of Munich / DLR (Hu et al.), OpenStreetMap contributors, converted to TACO. Pixel values and labels are kept as released except where the description below says otherwise. All credit belongs to the original authors: if you use it, please cite them and follow their licence.

original dataset · paper · licence: CC-BY-SA-4.0, ODbL-1.0

Repackaged into TACO by the Image and Signal Processing Group (ISP), Universitat de València, within the ELLIOT project.

Citation

Please cite the original work:

@article{hu2023mdas,
  title   = {MDAS: a new multimodal benchmark dataset for remote sensing},
  author  = {Hu, Jingliang and Liu, Rong and Hong, Danfeng and Camero, Andr{\'e}s and Yao, Jing and Schneider, Mathias and Kurz, Franz and Segl, Karl and Zhu, Xiao Xiang},
  journal = {Earth System Science Data},
  volume  = {15},
  number  = {1},
  pages   = {113--131},
  year    = {2023},
  doi     = {10.5194/essd-15-113-2023}
}

About the data

Three co-registered sub-areas of Augsburg, Germany, acquired 7 May 2018. Converted from the original GeoTIFFs without resampling: every leaf keeps its native grid.

90 samples · splits: test 30 · train 30 · validation 30 · tasks: semantic-segmentation, super-resolution

Packaged as TACO v3.

Full description

Products. Eight sensor products at six ground sampling distances:

• 0.2 m 3K aerial RGB and 0.3 m 3K DSM
• 2.2 m HySpex hyperspectral, 368 bands
• 10 m and 30 m EeteS-simulated EnMAP, 242 bands
• 10 m EeteS-simulated Sentinel-2, Sentinel-2 L2A (12 bands) and Sentinel-1 GRD

Three OpenStreetMap reference rasters come with them: 19-class land use, 5-class water and binary buildings.

Bands. Hyperspectral channel centres and FWHM are taken from the release's own ENVI headers.

Omissions. The 2.2 m EeteS EnMAP product exists for sub_area_1 only and is not carried. The 3K RGB's constant alpha channel is dropped.

Legends. Land use keeps its native legend with no canonical mapping, being land use rather than land cover. Water is additionally mapped onto ml:lc11.

Splits. The release publishes none. Each sub-area is cut into 30 tiles, and split assigns whole sub-areas: sub_area_3 is train, sub_area_1 validation and sub_area_2 test.

Getting started

git clone https://github.com/OscarPellicer/taco
pip install -e "taco/python[ml]"     # builds the native reader: needs CMake and Ninja

Read it straight from the Hub:

import os
from huggingface_hub import hf_hub_download, snapshot_download
from taco.ml import Dataset, plot_sample

path = hf_hub_download("isp-uv-es/mdas-taco", "mdas.zip", repo_type="dataset")
ds = Dataset(path)
plot_sample(ds[0])

or from a local copy:

ds = Dataset("mdas.zip")
sample = ds[0]                      # {slot name: SlotValue}, arrays decoded
sample["rgb_3k"].array.shape

Metadata without decoding anything:

import taco
taco.read("mdas.zip")              # one Arrow table, levels joined

Samples

What a sample contains

role	slot	holds	modality	detail
input	rgb_3k	raster	optical	3 band(s), requantised
input	dsm_3k	raster	elevation	1 band(s), unit m, physical
input	hyspex	raster	hyperspectral	368 band(s), unit 1, scaled
input	enmap_10m	raster	hyperspectral	242 band(s), unit 1, scaled
input	enmap_30m	raster	hyperspectral	242 band(s), unit 1, scaled
input	eetes_s2_10m	raster	optical	4 band(s), unit 1, scaled
input	s2	raster	multispectral	12 band(s), unit 1, scaled
input	s1	raster	sar	2 band(s), unit 1, physical
input	rgb_3k_valid	mask	label_raster	2 classes
target	land_use	mask	label_raster	19 classes
target	water	mask	label_raster	5 classes
target	buildings	mask	label_raster	2 classes

Licence

CC-BY-SA-4.0, ODbL-1.0

Terms of use:

• Adapted material must be shared under CC-BY-SA-4.0.
• The land-use, water and building layers are derived from OpenStreetMap; a database adapted from them may be shared only under ODbL-1.0.

Required credits:

• MDAS: Hu, Liu, Hong et al. (TUM / DLR)
• © OpenStreetMap contributors (ODbL)
• Contains modified Copernicus Sentinel data

Providers: Technical University of Munich / DLR (Hu et al.), OpenStreetMap contributors

Acknowledgements

TACO was designed by César Aybar and is specified at https://asterisk.coop/taco/spec/.

Built by Oscar Pellicer within the ELLIOT project at the Image and Signal Processing Group (ISP), Universitat de València.