/datasets-1

Datasets for deep learning with satellite & aerial imagery

Datasets for deep learning applied to satellite and aerial imagery.

How to use this repository: if you know exactly what you are looking for (e.g. you have the paper name) you can Control+F to search for it in this page (or search in the raw markdown).

Lists of datasets

Remote sensing dataset hubs

Sentinel

As part of the EU Copernicus program, multiple Sentinel satellites are capturing imagery -> see wikipedia

Landsat

Long running US program -> see Wikipedia

Maxar

Satellites owned by Maxar (formerly DigitalGlobe) include GeoEye-1, WorldView-2, 3 & 4

  • Maxar Open Data Program provides pre and post-event high-resolution satellite imagery in support of emergency planning, response, damage assessment, and recovery
  • WorldView-2 European Cities -> dataset covering the most populated areas in Europe at 40 cm resolution

Planet

UC Merced

Land use classification dataset with 21 classes and 100 RGB TIFF images for each class. Each image measures 256x256 pixels with a pixel resolution of 1 foot

EuroSAT

Land use classification dataset of Sentinel-2 satellite images covering 13 spectral bands and consisting of 10 classes with 27000 labeled and geo-referenced samples. Available in RGB and 13 band versions

PatternNet

Land use classification dataset with 38 classes and 800 RGB JPG images for each class

Gaofen Image Dataset (GID) for classification

Million-AID

A large-scale benchmark dataset containing million instances for RS scene classification, 51 scene categories organized by the hierarchical category

DIOR object detection dataset

A large-scale benchmark dataset for object detection in optical remote sensing images, which consists of 23,463 images and 192,518 object instances annotated with horizontal bounding boxes

Multiscene

MultiScene dataset aims at two tasks: Developing algorithms for multi-scene recognition & Network learning with noisy labels

FAIR1M object detection dataset

A Benchmark Dataset for Fine-grained Object Recognition in High-Resolution Remote Sensing Imagery

  • arxiv papr
  • Download at gaofen-challenge.com
  • 2020Gaofen -> 2020 Gaofen Challenge data, baselines, and metrics

DOTA object detection dataset

A Large-Scale Benchmark and Challenges for Object Detection in Aerial Images. Segmentation annotations available in iSAID dataset

iSAID instance segmentation dataset

A Large-scale Dataset for Instance Segmentation in Aerial Images

HRSC RGB ship object detection dataset

SAR Ship Detection Dataset (SSDD)

High-Resolution SAR Rotation Ship Detection Dataset (SRSDD)

LEVIR ship dataset

A dataset for tiny ship detection under medium-resolution remote sensing images. Annotations in bounding box format

SAR Aircraft Detection Dataset

2966 non-overlapped 224×224 slices are collected with 7835 aircraft targets

xView1: Objects in context for overhead imagery

A fine-grained object detection dataset with 60 object classes along an ontology of 8 class types. Over 1,000,000 objects across over 1,400 km^2 of 0.3m resolution imagery. Annotations in bounding box format

xView2: xBD building damage assessment

Annotated high-resolution satellite imagery for building damage assessment, precise segmentation masks and damage labels on a four-level spectrum, 0.3m resolution imagery

xView3: Detecting dark vessels in SAR

Detecting dark vessels engaged in illegal, unreported, and unregulated (IUU) fishing activities on synthetic aperture radar (SAR) imagery. With human and algorithm annotated instances of vessels and fixed infrastructure across 43,200,000 km^2 of Sentinel-1 imagery, this multi-modal dataset enables algorithms to detect and classify dark vessels

Vehicle Detection in Aerial Imagery (VEDAI)

Vehicle Detection in Aerial Imagery. Bounding box annotations

Cars Overhead With Context (COWC)

Large set of annotated cars from overhead. Established baseline for object detection and counting tasks. Annotations in bounding box format

AI-TOD - tiny object detection

The mean size of objects in AI-TOD is about 12.8 pixels, which is much smaller than other datasets. Annotations in bounding box format

RarePlanes

Counting from Sky

A Large-scale Dataset for Remote Sensing Object Counting and A Benchmark Method

AIRS (Aerial Imagery for Roof Segmentation)

Public dataset for roof segmentation from very-high-resolution aerial imagery (7.5cm). Covers almost the full area of Christchurch, the largest city in the South Island of New Zealand.

Inria building/not building segmentation dataset

RGB GeoTIFF at spatial resolution of 0.3 m. Data covering Austin, Chicago, Kitsap County, Western & Easter Tyrol, Innsbruck, San Francisco & Vienna

AICrowd Mapping Challenge: building segmentation dataset

300x300 pixel RGB images with annotations in COCO format. Imagery appears to be global but with significant fraction from North America

  • Dataset release as part of the mapping-challenge
  • Winning solution published by neptune.ai here, achieved precision 0.943 and recall 0.954 using Unet with Resnet.
  • mappingchallenge -> YOLOv5 applied to the AICrowd Mapping Challenge dataset

BONAI - building footprint dataset

BONAI (Buildings in Off-Nadir Aerial Images) is a dataset for building footprint extraction (BFE) in off-nadir aerial images

GID15 large scale semantic segmentation dataset

LEVIR-CD building change detection dataset

Onera Sentinel-2 change detection dataset

It comprises 24 pairs of multispectral images taken from the Sentinel-2 satellites between 2015 and 2018.

SECOND - semantic change detection

ISPRS

Semantic segmentation dataset. 38 patches of 6000x6000 pixels, each consisting of a true orthophoto (TOP) extracted from a larger TOP mosaic, and a DSM. Resolution 5 cm

SpaceNet

SpaceNet is a series of competitions with datasets and utilities provided. The challenges covered are: (1 & 2) building segmentation, (3) road segmentation, (4) off-nadir buildings, (5) road network extraction, (6) multi-senor mapping, (7) multi-temporal urban change, (8) Flood Detection Challenge Using Multiclass Segmentation

WorldStrat Dataset

Nearly 10,000 km² of free high-resolution satellite imagery of unique locations which ensure stratified representation of all types of land-use across the world: from agriculture to ice caps, from forests to multiple urbanization densities.

Satlas

A Large-Scale, Multi-Task Dataset for Remote Sensing Image Understanding. Annotates all modalities (classification, segmentation, object detection etc)

  • Website
  • Dataset release in January 2023

FLAIR

  • https://ignf.github.io/FLAIR/
  • The FLAIR-one semantic segmentation dataset consists of 77,412 high resolution patches (512x512 at 0.2 m spatial resolution) with 19 semantic classes

RF100: object detection benchmark

RF100 is compiled from 100 real world datasets that straddle a range of domains. The aim is that performance evaluation on this dataset will enable a more nuanced guide of how a model will perform in different domains. Contains 10k aerial images

Tensorflow datasets

  • resisc45 -> RESISC45 dataset is a publicly available benchmark for Remote Sensing Image Scene Classification (RESISC), created by Northwestern Polytechnical University (NWPU). This dataset contains 31,500 images, covering 45 scene classes with 700 images in each class.
  • eurosat -> EuroSAT dataset is based on Sentinel-2 satellite images covering 13 spectral bands and consisting of 10 classes with 27000 labeled and geo-referenced samples.
  • BigEarthNet -> a large-scale Sentinel-2 land use classification dataset, consisting of 590,326 Sentinel-2 image patches. The image patch size on the ground is 1.2 x 1.2 km with variable image size depending on the channel resolution. This is a multi-label dataset with 43 imbalanced labels. Official website includes version of the dataset with Sentinel 1 & 2 chips
  • so2sat -> a dataset consisting of co-registered synthetic aperture radar and multispectral optical image patches acquired by Sentinel 1 & 2

Microsoft datasets

Google datasets

Google Earth Engine (GEE)

Since there is a whole community around GEE I will not reproduce it here but list very select references. Get started at https://developers.google.com/earth-engine/

Image captioning datasets

Weather Datasets

Cloud datasets

Forest datasets

Geospatial datasets

  • Resource Watch provides a wide range of geospatial datasets and a UI to visualise them

Time series & change detection datasets

  • BreizhCrops -> A Time Series Dataset for Crop Type Mapping
  • The SeCo dataset contains image patches from Sentinel-2 tiles captured at different timestamps at each geographical location. Download SeCo here
  • SYSU-CD -> The dataset contains 20000 pairs of 0.5-m aerial images of size 256×256 taken between the years 2007 and 2014 in Hong Kong

DEM (digital elevation maps)

  • Shuttle Radar Topography Mission, search online at usgs.gov
  • Copernicus Digital Elevation Model (DEM) on S3, represents the surface of the Earth including buildings, infrastructure and vegetation. Data is provided as Cloud Optimized GeoTIFFs. link
  • Awesome-DEM

UAV & Drone datasets

Other datasets

  • land-use-land-cover-datasets
  • EORSSD-dataset -> Extended Optical Remote Sensing Saliency Detection (EORSSD) Dataset
  • RSD46-WHU -> 46 scene classes for image classification, free for education, research and commercial use
  • RSOD-Dataset -> dataset for object detection in PASCAL VOC format. Aircraft, playgrounds, overpasses & oiltanks
  • VHR-10_dataset_coco -> Object detection and instance segmentation dataset based on NWPU VHR-10 dataset. RGB & SAR
  • HRSID -> high resolution sar images dataset for ship detection, semantic segmentation, and instance segmentation tasks
  • MAR20 -> Military Aircraft Recognition dataset
  • RSSCN7 -> Dataset of the article “Deep Learning Based Feature Selection for Remote Sensing Scene Classification”
  • Sewage-Treatment-Plant-Dataset -> object detection
  • TGRS-HRRSD-Dataset -> High Resolution Remote Sensing Detection (HRRSD)
  • MUSIC4HA -> MUltiband Satellite Imagery for object Classification (MUSIC) to detect Hot Area
  • MUSIC4GC -> MUltiband Satellite Imagery for object Classification (MUSIC) to detect Golf Course
  • MUSIC4P3 -> MUltiband Satellite Imagery for object Classification (MUSIC) to detect Photovoltaic Power Plants (solar panels)
  • ABCDdataset -> damage detection dataset to identify whether buildings have been washed-away by tsunami
  • OGST -> Oil and Gas Tank Dataset
  • LS-SSDD-v1.0-OPEN -> Large-Scale SAR Ship Detection Dataset
  • S2Looking -> A Satellite Side-Looking Dataset for Building Change Detection, paper
  • Zurich Summer Dataset -> Semantic segmentation of urban scenes
  • AISD -> Aerial Imagery dataset for Shadow Detection
  • Awesome-Remote-Sensing-Relative-Radiometric-Normalization-Datasets
  • SearchAndRescueNet -> Satellite Imagery for Search And Rescue Dataset, with example Faster R-CNN model
  • geonrw -> orthorectified aerial photographs, LiDAR derived digital elevation models and segmentation maps with 10 classes. With repo
  • Thermal power plans dataset
  • University1652-Baseline -> A Multi-view Multi-source Benchmark for Drone-based Geo-localization
  • benchmark_ISPRS2021 -> A new stereo dense matching benchmark dataset for deep learning
  • WHU-SEN-City -> A paired SAR-to-optical image translation dataset which covers 34 big cities of China
  • SAR_vehicle_detection_dataset -> 104 SAR images for vehicle detection, collected from Sandia MiniSAR/FARAD SAR images and MSTAR images
  • ERA-DATASET -> A Dataset and Deep Learning Benchmark for Event Recognition in Aerial Videos
  • SSL4EO-S12 -> a large-scale dataset for self-supervised learning in Earth observation
  • UBC-dataset -> a dataset for building detection and classification from very high-resolution satellite imagery with the focus on object-level interpretation of individual buildings
  • AIR-CD -> a challenging cloud detection data set called AIR-CD, with higher spatial resolution and more representative landcover types
  • AIR-PolSAR-Seg -> a challenging PolSAR terrain segmentation dataset
  • HRC_WHU -> High-Resolution Cloud Detection Dataset comprising 150 RGB images and a resolution varying from 0.5 to 15 m in different global regions
  • AeroRIT -> A New Scene for Hyperspectral Image Analysis
  • Building_Dataset -> High-speed Rail Line Building Dataset Display
  • Haiming-Z/MtS-WH-reference-map -> a reference map for change detection based on MtS-WH
  • MtS-WH-Dataset -> Multi-temporal Scene WuHan (MtS-WH) Dataset
  • Multi-modality-image-matching -> image matching dataset including several remote sensing modalities
  • RID -> Roof Information Dataset for CV-Based Photovoltaic Potential Assessment. With paper
  • APKLOT -> A dataset for aerial parking block segmentation
  • QXS-SAROPT -> Optical and SAR pairing dataset from the paper: The QXS-SAROPT Dataset for Deep Learning in SAR-Optical Data Fusion
  • SAR-ACD -> SAR-ACD consists of 4322 aircraft clips with 6 civil aircraft categories and 14 other aircraft categories
  • SODA -> A large-scale Small Object Detection dataset. SODA-A comprises 2510 high-resolution images of aerial scenes, which has 800203 instances annotated with oriented rectangle box annotations over 9 classes.
  • Data-CSHSI -> Open source datasets for Cross-Scene Hyperspectral Image Classification, includes Houston, Pavia & HyRank datasets
  • SynthWakeSAR -> A Synthetic SAR Dataset for Deep Learning Classification of Ships at Sea, with paper
  • SAR2Opt-Heterogeneous-Dataset -> SAR-optical images to be used as a benchmark in change detection and image transaltion on remote sensing images
  • urban-tree-detection-data -> Dataset for training and evaluating tree detectors in urban environments with aerial imagery
  • Landsat 8 Cloud Cover Assessment Validation Data
  • Attribute-Cooperated-Classification-Datasets -> Three datasets based on AID, UCM, and Sydney. For each image, there is a label of scene classification and a label vector of attribute items.
  • dynnet -> DynamicEarthNet: Daily Multi-Spectral Satellite Dataset for Semantic Change Segmentation
  • open_earth_map -> a benchmark dataset for global high-resolution land cover mapping
  • Satellite imagery datasets containing ships -> A list of radar and optical satellite datasets for ship detection, classification, semantic segmentation and instance segmentation tasks
  • SolarDK -> A high-resolution urban solar panel image classification and localization dataset
  • Roofline-Extraction -> dataset for paper 'Knowledge-Based 3D Building Reconstruction (3DBR) Using Single Aerial Images and Convolutional Neural Networks (CNNs)'
  • Building-detection-and-roof-type-recognition -> datasets for the paper 'A CNN-Based Approach for Automatic Building Detection and Recognition of Roof Types Using a Single Aerial Image'
  • PanCollection -> Pansharpening Datasets from WorldView 2, WorldView 3, QuickBird, Gaofen 2 sensors
  • OnlyPlanes -> Synthetic dataset and pretrained models for Detectron2
  • Remote Sensing Satellite Video Dataset for Super-resolution
  • WHU-Stereo -> A Challenging Benchmark for Stereo Matching of High-Resolution Satellite Images
  • BH-POOLS & BH-WATERTANKS -> segmentation dataset of swimming pools and water tanks in Brazil
  • BrazilDAM Dataset -> a multi sensor (Landsat 8 and Sentinel 2) and multitemporal dataset that consists of multispectral images of ore tailings dams throughout Brazil
  • Bridge Dataset -> 500 images each containing at least one bridge
  • Brazilian Cerrado-Savanna Scenes Dataset -> 1,311 multi-spectral scenes extracted from images acquired by the RapidEye are partitioned into 4 classes: Agriculture, Arboreal, Herbaceous and Shrubby Vegetation
  • Brazilian Coffee Scenes Dataset
  • FireRisk -> A Remote Sensing Dataset for Fire Risk Assessment with Benchmarks Using Supervised and Self-supervised Learning
  • Road-Change-Detection-Dataset
  • 3DCD -> infer 3D CD maps using only remote sensing optical bitemporal images as input without the need of Digital Elevation Models (DEMs)
  • Hyperspectral Change Detection Dataset Irrigated Agricultural Area
  • CNN-RNN-Yield-Prediction -> soybean dataset

Kaggle

Kaggle hosts over > 200 satellite image datasets, search results here. The kaggle blog is an interesting read.

Kaggle - Amazon from space - classification challenge

Kaggle - DSTL segmentation challenge

Kaggle - DeepSat land cover classification

Kaggle - Airbus ship detection challenge

Kaggle - Shipsnet classification dataset

Kaggle - Ships in Google Earth

Kaggle - Ships in San Franciso Bay

Kaggle - Swimming pool and car detection using satellite imagery

Kaggle - Planesnet classification dataset

Kaggle - CGI Planes in Satellite Imagery w/ BBoxes

Kaggle - Draper challenge to place images in order of time

Kaggle - Dubai segmentation

Kaggle - Massachusetts Roads & Buildings Datasets - segmentation

Kaggle - Deepsat classification challenge

Not satellite but airborne imagery. Each sample image is 28x28 pixels and consists of 4 bands - red, green, blue and near infrared. The training and test labels are one-hot encoded 1x6 vectors. Each image patch is size normalized to 28x28 pixels. Data in .mat Matlab format. JPEG?

  • Sat4 500,000 image patches covering four broad land cover classes - barren land, trees, grassland and a class that consists of all land cover classes other than the above three
  • Sat6 405,000 image patches each of size 28x28 and covering 6 landcover classes - barren land, trees, grassland, roads, buildings and water bodies.

Kaggle - High resolution ship collections 2016 (HRSC2016)

Kaggle - SWIM-Ship Wake Imagery Mass

Kaggle - Understanding Clouds from Satellite Images

In this challenge, you will build a model to classify cloud organization patterns from satellite images.

Kaggle - 38-Cloud Cloud Segmentation

Kaggle - Airbus Aircraft Detection Dataset

Kaggle - Airbus oil storage detection dataset

Kaggle - Satellite images of hurricane damage

Kaggle - Austin Zoning Satellite Images

Kaggle - Statoil/C-CORE Iceberg Classifier Challenge

Classify the target in a SAR image chip as either a ship or an iceberg. The dataset for the competition included 5000 images extracted from multichannel SAR data collected by the Sentinel-1 satellite. Top entries used ensembles to boost prediction accuracy from about 92% to 97%.

Kaggle - Land Cover Classification Dataset from DeepGlobe Challenge - segmentation

Kaggle - Next Day Wildfire Spread

A Data Set to Predict Wildfire Spreading from Remote-Sensing Data

Kaggle - Satellite Next Day Wildfire Spread

Inspired by the above dataset, using different data sources

Kaggle - Spacenet 7 Multi-Temporal Urban Change Detection

Kaggle - Satellite Images to predict poverty in Africa

Kaggle - NOAA Fisheries Steller Sea Lion Population Count

Kaggle - Arctic Sea Ice Image Masking

Kaggle - Overhead-MNIST

Kaggle - Satellite Image Classification

Kaggle - EuroSAT - Sentinel-2 Dataset

Kaggle - Satellite Images of Water Bodies

Kaggle - NOAA sea lion count

Kaggle - miscellaneous

Competitions

Competitions are an excellent source for accessing clean, ready-to-use satellite datasets and model benchmarks.