Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for machine learning. One of the methods includes receiving one or more page images from a document; for each page image: providing the page image to a computer vision neural network model, wherein the neural network model is trained for the particular page type and is configured to output predictions of coordinates for one or more regions within the image and corresponding labels for the respective regions; and generating an output data structure associating each labeled region with text content located within the labeled region.