Links to related benchmarks

Here you can find a list of publicly available benchmarks involving machine learning and computer vision tasks on a moderate to large-scale geospatial datasets: 

• ISPRS Benchmarks: semantic labeling. Aerial multi-spectral image data and elevation information.
• DeepGlobe: High-res satellite imagery, 3 tasks: road extraction, building detection, land-cover segmentation
• UNESCO / EPFL Mapping Challenge: building instance segmentation. Aerial RGB data
• IEEE 2015 Data Fusion contest (complete): semantic labeling. Aerial multi-spectral image data, color and 3D point cloud.
• IEEE 2017 Data Fusion contest (complete): local climate-zone prediction. Satellite multi-spectral and multi-temporal image data and vector layers.
• DigitalGlobe SpaceNet, round 2 (live now). Large-scale building footprint extraction. Satellite multi-spectral image data.
• DSTL object detection challenge (kaggle, complete). Satellite multi-spectral image data.
• INRIA aerial image labeling dataset: building segmentation. Aerial image data.
• Semantic3D: Large-scale semantic labeling of 3D point clouds. Ground-level lidar.
• UC Merced dataset: tile-based land-use classification. Satellite image data.
• Zurich Summer Dataset: Semantic segmentation with scarce labels. Satellite acquisitions pansharpened QuickBird.
• Brasilian Coffee plantation dataset: coffee crop classification. Satellite image data.
• Cityscapes: Stereo video sequences of  street-level scenes. Pixel-level annotations.
• Toronto City Dataset (coming soon): semantic labeling, 3D building reconstruction, road centerline extraction, and others. Aerial image data, ground level imagery, panoramas, and other multi-source information.
• Kitti: Stereo video sequences of street-level scenes, 3D point clouds and GPS. Object-level annotations (cars and pedestrians).
• ApolloScape: Stereo videos sequencies of street-level scenes, 3D point clouds and panoramic images. Pixel level annotations

If you think your dataset belongs to this list, please contact us!