Digital image search training techniques and machine-learning architectures are described. In one example, a query digital image is received by service provider system, which is then used to select at least one positive sample digital image, e.g., having a same product ID. A plurality of negative sample digital images is also selected by the service provider system based on the query digital image, e.g., having different product IDs. The at least one positive sample digital image and the plurality of negative samples are then aggregated by the service provider system into a single aggregated digital image. At least one neural network is then trained by the service provider system using a loss function based on a feature comparison between the query digital image and samples from the aggregated digital image in a single pass.

The disclosed computer-implemented method may include (i) identifying, by a server computer system, a provider computing device for use in capturing street-level image data, where the provider computing device controls a camera positioned to capture street-level imagery outside the vehicle, (ii) determining, by the server computer system, a configuration that controls use of the provider computing device to provide street-level image data captured by the camera to the server computer system, (iii) sending, by the server computer system, the configuration to the computing device, and (iv) receiving, from the computing device, street-level image data captured by the computing device using the camera responsive to the configuration. Various other methods, systems, and computer-readable media are also disclosed.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, are described for resolving duplicate images. In one aspect, a method includes obtaining a selection of a single image from among a plurality of duplicate images. Each duplicate image has an associated set of metadata. The method also includes aggregating each set of metadata into aggregated information, and storing the selected image together with the aggregated information on data storage accessible to a data processing apparatus.

Compositing aware digital image search techniques and systems are described that leverage machine learning. In one example, a compositing aware image search system employs a two-stream convolutional neural network (CNN) to jointly learn feature embeddings from foreground digital images that capture a foreground object and background digital images that capture a background scene. In order to train models of the convolutional neural networks, triplets of training digital images are used. Each triplet may include a positive foreground digital image and a positive background digital image taken from the same digital image. The triplet also contains a negative foreground or background digital image that is dissimilar to the positive foreground or background digital image that is also included as part of the triplet.

One or more computing devices, systems, and/or methods are provided. A first image captured via a first camera is received. The first image is analyzed to identify a first object within the first image. An object tag comprising information associated with the first object is generated. The object tag and/or object information associated with the first object are stored. A second image captured via a second camera is received. The first object is identified within the second image based upon the second image and/or the object information. A representation of the object tag may be displayed via a display device. Alternatively and/or additionally, a location of the first object may be determined based upon the second image. Alternatively and/or additionally, an audio message indicative of the object tag may be output via a speaker.

An image recognition approach employs both computer generated and manual image reviews to generate image tags characterizing an image. The computer generated and manual image reviews can be performed sequentially or in parallel. The generated image tags may be provided to a requester in real-time, be used to select an advertisement, and/or be used as the basis of an internet search. In some embodiments generated image tags are used as a basis for an upgraded image review. A confidence of a computer generated image review may be used to determine whether or not to perform a manual image review.

Various methods for generating a content corpus populated with content related to a particular geographic area are provided herein. One example method comprises, for each document in an initial local content corpus, applying a first set of heuristic filters to the raw content of each document, identifying at least a second term, applying a second set of heuristic filters to the raw content of each document, the second set of heuristic filters associated with the second term, iteratively performing the identification of additional terms and application of an additional set of heuristic filters associated with the additional terms until each identifiable term is extracted, determining a level on a geographic containment hierarchy indicative of a location to which each document from the set of documents is local, and for each place in a gazette, and for each document, determining a set of points in polygons indicative of its locality.

Various methods for generating a content corpus populated with content related to a particular geographic area are provided herein. One example method comprises, for each document in an initial local content corpus, applying a first set of heuristic filters to the raw content of each document, identifying at least a second term, applying a second set of heuristic filters to the raw content of each document, the second set of heuristic filters associated with the second term, iteratively performing the identification of additional terms and application of an additional set of heuristic filters associated with the additional terms until each identifiable term is extracted, determining a level on a geographic containment hierarchy indicative of a location to which each document from the set of documents is local, and for each place in a gazette, and for each document, determining a set of points in polygons indicative of its locality.

A system and method for geolocation-based pictographs are provided. In example embodiments, a current geolocation of a user device is determined. A pictograph is identified based on the current geolocation of the user device. The identified pictograph is presented on a user interface of the user device.

A system and method for geolocation-based pictographs are provided. In example embodiments, a current geolocation of a user device is determined. A pictograph is identified based on the current geolocation of the user device. The identified pictograph is presented on a user interface of the user device.