Data & Annotation Team

Expertise

Development Seed’s Data and Annotation Team produces high-quality geodata and machine learning labels.

High-quality data is crucial in Geospatial and Machine learning fields for making any project successful.

Why it matters

Our Data Team is one of the most prolific and accomplished mapping teams in the world. We have extensive experience with OpenStreetMap, the open geospatial software ecosystem, and its metadata standards. We are also at the forefront of building open and highly efficient workflows for mapping and building complex machine learning training datasets. To illustrate, our team edited more than 25 million objects in OpenStreetMap so far.

  • Data quality

    Data quality is a fundamental aspect that improves the outcomes of our projects, so we combine advanced machine learning methods with highly-trained mappers to quickly and consistently produce pixel-perfect maps and imagery annotation.
  • Vastly speed in mapping and annotation

    We vastly speed up mapping through the use of machine learning. We have developed and refined detection algorithms for most common features including settlements and buildings, road and power infrastructure, and agriculture and land features. Aided by these algorithms, our Data and Annotation Team can work up to 30x faster.

Fast, high quality mapping and validation for the World Bank

We mapped the full power grid of Zambia, Nigeria, and Pakistan with the World Bank. Specifically, we added over 125k features (transformers, poles, switches, substations) and over 28k kilometers of power lines to OpenStreetMap.

High quality training data

We produced image classification, object detection and semantic segmentation training datasets for our ML projects.

Creating over 15,000 annotations for building construction quality, structure, and materials used for training ML model for categorizing buildings according to risk.

Labeling schools to apply scalable machine learning over high-resolution satellite imagery and validating the ML inferences to map every school in several countries from Africa, Asia and South America.

Efficient workflows for high speed mapping

We build and customize a variety of tools in order to maximize speed while maintaining extremely high accuracy in our data generation and validations workflows. Some of our main tools are Chips-ahoy and Relabeler, for ML output validations; Geokit, for data analysis and processing; and osm-seed tool that provides an easily installable package for the OpenStreetMap software stack. You can read more about our tools and workflows.

Pricing for mapping

COMPONENT COST IN USD
Building
Low density mapping km² $23.00
Intermediate density mapping km² $138.00
High density mapping km² $344.00
Roads
Low density mapping km² $5.00
Intermediate density mapping km² $9.00
High density mapping km² $9.00
Farm Field
Low Complexity per farm $0.58
Medium Complexity per farm $0.88
High Complexity per farm $1.75

Pricing for annotation

Annotation type condition average Cost in USD
Classification 1 class per image 1,000 images/hour $35.00
Object detection 3 classes per image 80 images/hour $35.00