Here is presented the list of open datasets created by Aeronetlab group at Skoltech for objects recognition in satellite and aerial images. Most of datasets are distributed under the Open License within a single pipeline supported by a data access tools (check for Aeronetlib in our github page). These experimental datasets are to be used in training or validation of the deep learning algorithms.

Despite of increasing number of datasets and competitions in remote sensing data science and some large datasets that'd been provided to the research community (e.g. Spacenet) there is still a lack of geographical diversity and the number of training classes. The dataset is proposed to be extended to different data sources, territories and application domains in accordance with classification of the natural and man-made objects that have a clear interpretation either in satellite or in aerial imagery (see "markup classes").

Datasets

Project name	Number of datasets	Description	Size	Dwonload link
"Emergenсy mapping"	2	Emergency Mapping is a deeplearning method to detect destroyed (damaged) buildings in remote sensing imagery	235 Mb + 5.2 Gb	Download
"Building heights"	1	For validation of buldings heights using the method for heights reconstruction in single image by the known sun and satellite angles	1.2 Gb	Download

Markup classes

Classification for obects labeling in imagery (all classes)

ID	CLASS_NAME	Description	Visual	Application domains
0	clutter
Buildings and Construction
101	Residential building	Roofs (not footprints!) of apartment buildings. Shoud have 5+ storeys.		retail, real estate, urban, mapping
901	building shadow	Shadows of multi-storey buildings. Should be related to appropiate building		retail, real estate, urban
902	building wall	Walls of multi-storey buildings		retail, real estate, urban
102	House			retail, real estate, urban
103	Industrial building	Plants, etc.		retail, urban
104	Commercial building	Usualy hard to define only by imagery		retail, real estate, urban
105	Other non-residential buildings	Garages, hangars, etc. - mostly small non-residential buildings
106	Construction site	The site wherer construction work is going		retail, construction, real estate
107	Construction building			construction
108	Pit			construction
109	Swimming Pool	Swimming pool at the private residential area. Shoud have a relation with private house		retail, marketing
110	Religious
Oil&Gaz
201	Oil storage facility			oil&gas
202	Oil well			oil&gas
203	Gas station			oil&gas, transport
204	Oil spill			oil&gas, ecology
Roads
301	Highway			transport
302	Track			transport
303	Footpath			transport, tourism
304	Railway			transport
305	Bridge			transport, marine
306	Parking lot			transport
308	Cross walk			urban, transport
307	Other road			transport
309	Highway and tracks	Union of 301 and 302 classes		transport
Transport
401	Train			transport
402	Truck			transport
403	Car			transport
404	Vessel			transport, marine
405	Airplane			transport
406	Other vehicle			transport
501	Dock			marine
502	Container			marine, transport
Vegetation
601	Tree
602	No leaf tree
603	Forest
604	Low forest	Low or coppice forest
605	Palm tree
606	Other tree
607	Shrub	Shrubland
608	Plough
609	Сrops
610	Lawn
611	Grassland
612	Other low vegetation
613	Woodlands
614	TSV	Union for 603, 604, 606, 607 and 613 classes
Water
614	River
615	Lake
616	Swamp
617	Other water body
618	Stone
619	Clay
620	Sand
621	Other soil
622	Sea
Power infrastructure
701	Solar panel
702	Power tower
703	Cell tower
Emergency and risk management
801	Destroyed building
802	Damaged building
803	Landfill
804	Flooded area	Flooded residential area whrere the water poses a threat to locals
805	Flooded residential area
806	Forest loss	Forest losse due to wildfires, loggings etc.
807	Forest growth	Forest growth inverse to forest losses
808	Changes of residental buildings

Credentials

"Open datasets" is the joint project of Skoltech and University of Innopolis, maintained by AeronetLab at Skoltech. The goal of the project is to provide research and developers community with training datasets and benchmarks to develop deep learning algorithms for Earth Observation data analysis.