A curated list of resources dedicated to deep learning / computer vision algorithms for mapping. The mapping problems include road network inference, building footprint extraction, etc. Any suggestions and pull requests are welcome.

• [2015-ICCV] Enhancing Road Maps by Parsing Aerial Images Around the World paper
• [2016-KDD] City-Scale Map Creation and Updating Using GPS Collections paper
• [2017-ICCV] DeepRoadMapper: Extracting Road Topology From Aerial Images paper code
• [2017-CVPRW] Joint Learning From Earth Observation and OpenStreetMap Data to Get Faster Better Semantic Maps paper
• [2018-ISPRS International Journal of Geo-Information] Generative Street Addresses from Satellite Imagery paper code
• [2018-CVPR, RoadTracer] RoadTracer: Automatic Extraction of Road Networks from Aerial Images paper homepage code
• [2018-CVPRW2018] D-LinkNet : LinkNet with Pretrained Encoder and Dilated Convolution for High Resolution Satellite Imagery Road Extraction paper code

• [2017-TPAMI] Learning Building Extraction in Aerial Scenes with Convolutional Networks paper
• [2018-CVPRW2018, TernausNetV2] TernausNetV2: Fully Convolutional Network for Instance Segmentation paper code
• [2018-CVPRW2018] Building Detection From Satellite Imagery Using Ensemble of Size-Specific Detectors paper
• [2018-CVPRW2018] Building Detection from Satellite Imagery Using a Composite Loss Function paper

• OSM data export tools from HOTOSM, tool

• [2017-ICCV, TorontoCity] TorontoCity: Seeing the World with a Million Eyes homepage paper
  • Data source: aerial RGB image, streetview panorama, GoPro, stereo image, street-view LIDAR, airborne LIDAR; Maps: buildings and roads, 3D buildings, property meta-data; Tasks: semantic segmentation, building height estimation, instance segmentation, road topology, zoning segmentation and classification.
  • A non-sharing dataset.
• [2018-arxiv, SpaceNet] SpaceNet: A Remote Sensing Dataset and Challenge Series paper
  • SpaceNet competition hosts several datasets on roads and buildings. For example, Challenge 3 - Las Vegas, Paris, Shanghai, Khartoum Road Extraction Challenge provides road-centerlines annotation of SpaceNet dataset. It uses the imagery from Challenge 2 (24,586 images of 302,701 building footprints), only tiled into 400m chips (resolution of 1300px x 1300px.
• [2018-CVPRW2018, DeepGlobe] DeepGlobe 2018: A Challenge to Parse the Earth through Satellite Images paper
  • Road extraction challenge: DigitalGlobe+Vivid images in Thailand, Indonesia, India with 50cm/pixel. Total of 8570 images (6226 training, 1243 validation, 1101 testing).
  • Building detection: the images are taken from 25,586 images of size 650px x 650px
  • Land cover classification: containing 1,146 satellite images of size 2448px × 2448px pixels in total.