Map feature extraction system for computer map visualizations

A non-transitory computer readable storage medium configured to store program code, the program code comprising instructions that, when executed by one or more processors, cause the one or more processors to:

• identifying, in a set of map tile vector data, a plurality of map tiles that comprise a particular map feature;
• rasterizing vector data of each map tile of the plurality of map tiles to generate respective mask images, each mask image comprising a segment of pixels distinguishing the respective instance of the particular map feature from other pixels of the mask image;
• identifying geo-reference data of each map tile of the plurality of map tiles;
• retrieving, for each identified geo-reference data of each map tile of the plurality of map tiles, a corresponding aerial image that captures a geography specified by the geo-reference data;
• pairing each mask image with the corresponding aerial image of the map tile from which the mask image was generated;
• training a machine learned model using the pairs of mask images and corresponding aerial images to determine a probability that a pixel in an aerial image corresponds to an instance of the particular map feature;
• receiving a particular aerial image comprising a plurality of pixels;
• predicting, for each pixel of a set of pixels of the plurality of pixels of the particular aerial image, a probability that the pixel corresponds to the particular map feature;
• identifying a subset of pixels of the set of pixels, each pixel in the subset of pixels having a predicted probability that is greater than or equal to a threshold probability of corresponding to the particular map feature;
• and
• determining a bounded geometry enclosing at least a portion of the identified subset of pixels.

A system comprising a computer processor for executing computer program instructions, and a non-transitory computer-readable storage medium storing computer program instructions executable by the processor to perform operations comprising:

• identifying, in a set of map tile vector data, a plurality of map tiles that comprise a particular map feature;
• rasterizing vector data of each map tile of the plurality of map tiles to generate respective mask images, each mask image comprising a segment of pixels distinguishing the respective instance of the particular map feature from other pixels of the mask image;
• identifying geo-reference data of each map tile of the plurality of map tiles;
• retrieving, for each identified geo-reference data of each map tile of the plurality of map tiles, a corresponding aerial image that captures a geography specified by the geo-reference data;
• pairing each mask image with the corresponding aerial image of the map tile from which the mask image was generated;
• training a machine learned model using the pairs of mask images and corresponding aerial images to determine a probability that a pixel in an aerial image corresponds to an instance of the particular map feature;
• receiving a particular aerial image comprising a plurality of pixels;
• predicting, for each pixel of a set of pixels of the plurality of pixels of the particular aerial image, a probability that the pixel corresponds to the particular map feature;
• identifying a subset of pixels of the set of pixels, each pixel in the subset of pixels having a predicted probability that is greater than or equal to a threshold probability of corresponding to the particular map feature;
• and
• determining a bounded geometry enclosing at least a portion of the identified subset of pixels.