Awesome Satellite Imagery Competitions

List of machine learning competitions for satellite imagery and remote sensing. Sorted by submission deadline.

• Spacenet Challenge - Round 4 (CosmiQ Works, DigitalGlobe, Radiant Solutions, AWS, Dec 2018)
  Building object detection with Off-Nadir imagery. 126k building polygons (Atlanta). 27 WorldView 2 images from 7-54 degrees off-nadir angle. Bi-cubicly resampled to same number of pixels in each tile despite courser native resolution due to high off-nadir angles.

• Open AI Challenge: Tanzania (WeRobotics & Wordlbank, Nov 2018)
  Building object detection & building condition classification (3 classes), RGB UAV imagery - Link to data

• Airbus Ship Detection Challenge (Airbus, Sep 2018)
  Ship Object Detection, 104k train / 88k test image chips, object pixel masks in run-length encoding format, Kaggle kernels.

• Agricultural Crop Cover Classification Challenge (CrowdANALYTIX, Jul 2018)
  Semantic Segmentation (2 main categories: Corn, Soybeans), Landsat 8 imagery (30m), USDA Cropland Data Layer as ground truth.

• DOTA: Large-scale Dataset for Object Detection in Aerial Images (Wuhan University et al.)
  Object Detection (15 categories), 188k instances, Google Earth image chips, Faster-RCNN baseline model (MXNet), DOTA development kit, Academic use only

• xView 2018 Detection Challenge (DIUx, Jul 2018)
  Object Detection (60 categories), 1 million instances, Worldview-3 imagery (0.3m), COCO data format, pre-trained Tensorflow and Pytorch baseline models

• CrowdAI Mapping Challenge (Humanity & Inclusion NGO, May 2018)
  Semantic/Instance Segmentation (buildings), RGB sat. imagery, COCO data format

• Open AI Challenge: Aerial Imagery of South Pacific Islands (WeRobotics & Worldbank, May 2018)
  Object Detection (4 tree species), Semantic Segmentation (2 road types), RGB UAV imagery (0.4/0.8m), multiple AOIs in Tonga

• DEEPGLOBE - 2018 Satellite Challange (CVPR, Apr 2018)
  3 challenge tracks: Road Extraction, Building Detection, Land cover classification

• IEEE Data Fusion Contest 2018 (IEEE, -Mar 2018)
  Land cover classification (20 categories) by fusing data three sources: Multispectral LiDAR, Hyperspectral (1m), RGB imagery (0.05m)

• Spacenet Challenge - Round 3 (CosmiQ Works, Radiant Solutions, NVIDIA, Feb 2018)
  Road Extraction, multiple city aois, 3(RGB)/8band Worldview-3 imagery (0.3m), SpaceNet Challenge Asset Library

• Statoil/C-CORE Iceberg Classifier Challenge (Statoil/C-CORE, Jan 2018)
  Image Recognition (Predict if image chip contains ship or iceberg), 2-band HH/HV polarization SAR imagery, Kaggle kernels

• Functional Map of the World Challenge (IARPA, Dec 2017)
  Object Detection (63 categories), 1 million instances, 4/8 band sat. imagery, COCO data format, baseline models

• Urban 3D Challenge (USSOCOM, Dec 2017)
  Building footprint detection, RGB orthophotos (0.5m), 3 cities, SpaceNet Challenge Asset Library

• NIST DSE Plant Identification with NEON Remote Sensing Data (inria.fr, Oct 2017)
  Extraction of tree position, species and crown parameters, hyperspectral (1m) & RGB imagery (0.25m), LiDAR point cloud and canopy height model

• Planet: Understanding the Amazon from Space (Planet, Jul 2017)
  Image recognition (Predict 1 of 13 land cover and 1 of 4 cloud condition labels per image chip), Amazonian rainforest, 4 band sat. imagery (RGB-NIR, 3-5m), Kaggle kernels

• TiSeLaC : Time Series Land Cover Classification Challenge (UMR TETIS, Jul 2017)
  Land cover time series classification (9 categories), Landsat-8 (30m, 23 images time series, 10 band features), Reunion island

• NOAA Fisheries Steller Sea Lion Population Count (NOAA, Jun 2017)
  Object Detection (5 sea lion categories), ~ 80k instances, ~ 1k aerial images, Kaggle kernels

• Spacenet Challenge - Round 2 (CosmiQ Works, Radiant Solutions, NVIDIA, May 2017)
  Building extraction, multiple city aois, 3/8band Worldview-3 imagery (0.3m), SpaceNet Challenge Asset Library

• DSTL Satellite Imagery Feature Detection challenge (Dstl, Feb 2017)
  Object Detection & Classification (10 categories). 3/16 band Worldview 3 imagery (0.3m - 7.5m), Kaggle kernels

• Spacenet Challenge - Round 1 (CosmiQ Works, Radiant Solutions, NVIDIA, Jan 2017)
  Building extraction, Rio de Janeiro, 3/8band Worldview-3 imagery (0.5m mosaic), SpaceNet Challenge Asset Library

• Multi-View Stereo 3D Mapping Challenge (IARPA, Nov 2016)
  Develop of a Multi-View Stereo (MVS) 3D mapping algorithm that can convert high-resolution satellite images to 3D point clouds

• Draper Satellite Image Chronology (Draper, Jun 2016)
  Predict the chronological order of images taken at the same locations over 5 days, Kaggle kernels

• Inria Aerial Image Labeling (inria.fr)
  Semantic Segmentation (buildings), RGB aerial imagery (0.3m), 5 cities