Training Data


22 Individual Datasets
1,298,610 Buildings Labeled
100,043 Training Chips

All of the datasets produced for this project are STAC compliant and hosted on Radiant MLHubThe training labels and associated imagery are released under a CC BY-N.C. 4.0 license.

Ramp training data all follows the same format and consists of two corresponding parts; very high resolution satellite and drone imagery in GeoTiff format and corresponding vector labels that overlay the imagery in GeoJSON format. These two components are combined and tiled out into 256 pixel by 256 pixel “chips” to become Ramp training data. 

Due to limitations of time and resource, not all datasets have undergone a thorough and extensive review through our QA/QC pipeline. As a result we devised a two-tier system making it clear which datasets were reviewed thoroughly (Tier 1, 100% eyes on) and which datasets were not reviewed as thoroughly (Tier 2, undergone some review). 

Muscat, OMAN

Tier 1

Tile Count: 2,891

Building Count: 30,652

Imagery: 40 cm sourced from Maxar ODP 10500100271BF800

Karnataka, INDIA

Tier 1

Tile Count: 6,288

Building Count: 51,335

Imagery: 40 cm sourced from Maxar ODP 104001002CA32300

Les Cayes, HAITI

Tier 1

Tile Count: 1,430

Building Count: 28,549

Imagery: 47 cm sourced from Maxar ODP 10300100A450A500

Mesopotamia, ST. VINCENT

Tier 1

Tile Count: 3,024

Building Count: 33,531

Imagery: 40 cm sourced from Maxar ODP 10500100236CC900

Muzuzu, Malawi

Tier 1

Tile Count: 3,357

Building Count: 92,918

Imagery: 45 cm sourced from Maxar ODP 10500100195A6700

Manjama, Sierra Leone

Tier 1

Tile Count: 4,681

Building Count: 60,951

Imagery: 30 cm sourced from Maxar ODP 10400100472E6700

Wa, GHANA

Tier 1

Tile Count: 7,622

Building Count: 68,189

Imagery: 32 cm sourced from Maxar ODP 1040010056B6FA00

Hpa-an, MYANMAR

Tier 1

Tile Count: 3,667

Building Count: 44,765

Imagery: 35 cm sourced from Maxar ODP 1040010033320500

Bentiu, SOUTH SUDAN

Tier 1

Tile Count: 1,789

Building Count: 22,396

Imagery: 35 cm sourced from Maxar ODP 104001004DAECE00

Accra, GHANA

Tier 1

Tile Count: 1,330

Building Count: 40,786

Imagery: 30 cm originally sourced from Open Cities AI Challenge

Dar es Salaam, TANZANIA

Tier 1

Tile Count: 566

Building Count: 40,786

Imagery: 30 cm originally sourced from Open Cities AI Challenge

Shanghai, CHINA

Tier 1

Tile Count: 3,574

Building Count: 7,118

Imagery: 30 cm originally sourced from the SpaceNet2 AI Challenge

Paris, FRANCE

Tier 1

Tile Count: 1,027

Building Count: 3,468

Imagery: 30 cm originally sourced from the SpaceNet2 AI Challenge

Dhaka, BANGLADESH

Tier 1

Tile Count: 11,905

Building Count: 189,057

Imagery: 30 cm sourced from Maxar ODP BG_Dhaka_19Q3_V0_R6C3

Barishal, BANGLADESH

Tier 1

Tile Count: 3,024

Building Count: 41,248

Imagery: 40 cm sourced from Maxar ODP 105001001597B00

Jashore, BANGLADESH

Tier 1

Tile Count: 7,310

Building Count: 80,050

Imagery: 35 cm sourced from Maxar ODP 104001003BA7C900

Cox's Bazar, BANGLADESH

Tier 1

Tile Count: 2,375

Building Count: 26,875

Imagery: 35 cm sourced from Maxar ODP 10400100546E3700

Chittagong, BANGLADESH

Tier 1

Tile Count: 5,229

Building Count: 38,096

Imagery: 40 cm sourced from Maxar ODP 105001001AC9800

Sylhet, BANGLADESH

Tier 2

Tile Count: 16,217

Building Count: 135,375

Imagery: 30 cm sourced from Maxar ODP 2022 Bangladesh Flooding Release

Ndjamena, CHAD

Tier 2

Tile Count: 3,044

Building Count: 124,208

Imagery: 45 cm sourced from Maxar ODP 10300100AA405C00

Lubumbashi, DRC

Tier 2

Tile Count: 8,498

Building Count: 148,459

Imagery: 30 cm sourced from Maxar ODP 1040010058041300

Nairobi, KENYA

Tier 2

Tile Count: 1,195

Building Count: 24,707

Imagery: 30 cm sourced from Maxar ODP

Creative Commons Disclosure

CC BY-NC 4.0 license

This work is licensed under a Creative Commons  Attribution-NonCommercial 4.0 International License. You are free to share this document, meaning copy and redistribute the material in any medium or format. You are also free to adapt this document, meaning remix, transform, and build upon the material. You may not use this work for commercial purposes. You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.

Suggested Citation: DevGlobal Partners. (2022). Replicable AI for Microplanning (ramp): Data Labeling Specifications. https://rampml.global/