Machine Learning image classification projects are characterized by reviewing large amounts of images looking for certain patterns, for instance: school patterns, vacant lots, crop types, and so on.
The Data team along with Machine Learning engineering have built workflow and tools to review large numbers of images in an easy way.
The statistics below demonstrate the efficiency of tools and the Data team at image classification.
|Classes per image||Images per hour||Min/Max image size in pixels|
|1||1,000||256x256 / 2048x2048|
|2||800||256x256 / 2048x2048|
|3||700||256x256 / 2048x2048|
|4||600||256x256 / 2048x2048|
Data team's image classification tools
Development Seed's Data team has built its own tools to address these types of challenges, which allow them to do quick reviews.
The list below shows the tools that the Data team uses for image classification.
Chips-ahoy is an open-source tool, used for validating machine learning chips. This tool is useful when the tiles for validation are scattered.
Relabeler is an open-source tool, used for validating map tiles chips. This tool is useful when the tiles for validation are together or close to each other.
Java OpenStreetMap editor
Java OpenStreetMap editor, known as JOSM is an open-source tool, this tool works with map tiles. JOSM allows the editions to be easier and faster with points, lines, and polygons compared with QGIS. The Data team uses JOSM together with osm-seed, also during years Development Seed engineers got experience customizing JOSM for their own purposes it means according to the project necessity.
Data team's classification projects
The following list shows the projects of image classification for Machine Learning training data.