Abstract: A computer-implemented process is disclosed for generating a labeled overhead image of a geographical area. A plurality of ground level images of the geographical area is retrieved. A ground level feature map is generated, via a ground level convolutional neural network, based on features extracted from the plurality of ground level images. An overhead image of the geographical area is also retrieved. A joint feature map is generated, via an overhead convolutional neural network based on the ground level feature map and features extracted from the plurality of ground level images. Geospatial function values at a plurality of pixels of the overhead image are estimated based on at least the joint feature map and the overhead image. The plurality of pixels of the overhead image is labeled according to the estimated geospatial function values.

Abstract: A computer-implemented process is disclosed for generating a labeled overhead image of a geographical area. A plurality of ground level images of the geographical area is retrieved. A ground level feature map is generated, via a ground level convolutional neural network, based on features extracted from the plurality of ground level images. An overhead image of the geographical area is also retrieved. A joint feature map is generated, via an overhead convolutional neural network based on the ground level feature map and features extracted from the plurality of ground level images. Geospatial function values at a plurality of pixels of the overhead image are estimated based on at least the joint feature map and the overhead image. The plurality of pixels of the overhead image is labeled according to the estimated geospatial function values.