The disclosure describes novel technology for inferring scenes from images. In one example, the technology includes a system that can determine partition regions from one or more factors that are independent of the image data, for an image depicting a scene; receive image data including pixels forming the image; classify pixels of the image into one or more pixel types based on one or more pixel-level features; determine, for each partition region, a set of pixel characteristic data describing a portion of the image included in the partition region based on the one or more pixel types of pixels in the partition region; and classify a scene of the image based on the set of pixel characteristic data of each of the partition regions.