Abstract: Various embodiments enable an image recognition system reduce the number image match candidates before running a full-fledged pair-wise match on all image match candidates. In order to accomplish this, each inventory image can be assigned to a group. For example, a title for a book sold by an electronic marketplace could be available in multiple languages, in multiple bindings, and the book could be available in print, audio book, or electronic book. Each one of these variations could be associated with its own similarly looking inventory image, each of which could be returned as a valid match to a query image for the book. Accordingly, the inventory images for these variations could be assigned to a group for the book and, instead of geometrically processing an image for each variation, the image match system can process a single image representing all of the variations.

Abstract: An advertising system can be configured to determine advertisements that are responsive to targeting segments included in advertisement requests without having to individually evaluate Boolean or logical targeting expressions associated with each advertisement. For example, the advertising system can be configured to determine, for advertisements, one or more terms that correspond to at least a portion of the respective targeting expressions of the advertisement. When an advertisement request is received, a determination can be made as to whether any of the segments (e.g., age group) associated with that request correspond to any of the terms that were derived from the advertisements. The advertisement system can then determine any advertisements that match the term(s) that correspond to the request.

Abstract: A semantic analysis can be performed to determine an intent of a received query. The intent can relate to a primary object of the query, which can be identified through the semantic analysis. Other attributes can be determined from the query that help to focus the object of the intent. A query vector is generated, based on the intent and primary object, and used to search a multi-dimensional semantic space including semantic representations of possible matches. The attributes are used to adjust the query vector in the semantic space. Objects having vectors ending proximate the query vector are identified as potential search results, with the distance from the query vector being used as a ranking mechanism. If refinement is needed, a dialog is used to obtain additional information from the user. Once results are obtained with sufficient confidence, results can be returned as search results for the query.

Abstract: Because not all companies register or obtain barcodes (or other encodings or identifiers) from a single, consistent source, and because different companies can be associated with the same barcode at different times, there can be multiple products associated with a specific barcode. If a user requests information associated with a specific barcode, information associated with the user or request can be used to attempt to determine the associated product if more than one matching product is uncovered. In some embodiments, a location of a user device at approximately a time of the request can be used to attempt to determine the appropriate product, such as if the user is in a store that only offers one of the potential products. Other information can be used as well, such as past locations, purchase history, viewing patterns, and the like.

Abstract: Large scale instance recognition is provided that can take advantage of channel-wise pooling. A received query image is processed to extract a set of features that can be used to generate a set of region proposals. The proposed regions of image data are processed using a trained classifier to classify the regions as object or non-object regions. Extracted features for the object regions are processed using feature correlation against extracted features for a set of object images, each representing a classified object. Matching tensors generated from the comparison are processed using a spatial verification network to determine match scores for the various object images with respect to a specific object region. The match scores are used to determine which objects, or types of objects, are represented in the query image. Information or content associated with the matching objects can be provided as part of a response.

Abstract: The processing of search queries for a customer using a set of resource can balance performance with cost, in order to ensure that the processing satisfies customer performance requirements while attempting to minimize the cost to the customer for obtaining that performance. In addition to dynamically updating the allocation of resources used to process the requests, such as to change the number or sizes of allocated resources, the number of indexes to be searched can be updated as well. For example, a search index can be divided into two sub-indexes against which queries can be executed concurrently in order to reduce the latency by about a half. Adjustments to the indexes and resources can be balanced to minimize cost while retaining performance, which can include rejoining sub-indexes or reducing the number of resource as appropriate.

Abstract: Memory management is provided for processors, such as GPUs used to process data using a trained machine learning model. Requests received to a CPU can be stored to a request queue until the queue is full, or until a timeout value has been reached for periods of lower activity. The requests can then be batched and sent to a GPU as a single message on a single thread. Memory can be pre-allocated, and the trained model loaded into GPU memory once for processing of the relevant batches. The individual requests can be processed by the GPU and the results analyzed to determine at least a subset of results to return to the CPU, which can be provided back as results of the processing.

Abstract: Content can be located for items that are stylistically similar to an item of interest. The item of interest can be represented in a query image, which is analyzed to determine one or more regions having an item represented therein. The classification of the item is determined, enabling identification of a trained model to be used to process image data for the region(s) of the query image. The trained model outputs a set of attributes, relating to visual or stylistic attributes, and corresponding confidence or prominence values for the attributes. These attributes and values can be compared against a data repository to locate items determined to be similar based on corresponding attributes and values. A similarity determination algorithm can identify similar items and rank those items by similarity. Content for the most similar items is returned as a result for the query image.

Abstract: Systems and methods for communicating in a network using share signals in accordance with various embodiments of the present disclosure are provided. In one embodiment, a method for sharing video footage captured by audio/video (A/V) recording and communication devices is provided, the method comprising: receiving from a first client device a share signal including first image data including a person of interest; receiving from a second A/V recording and communication device second image data; transmitting to a second client device, the first image data and a request to check the second image data; receiving from the second client device a locate person command signal; searching the second image data for a match of the person of interest; and when a match of the person of interest is found, generating and transmitting an alert signal that includes a notification that the person of interest was matched in the second image data.

Abstract: Approaches herein provide for multi-camera calibration and subsequent application in augmented reality applications. The approach obtains image data of a three-dimensional (3D) calibration object from different directions using the at least one camera. A determination is made for extrinsic parameters associated with a relative position of the 3D calibration object to the at least one camera. A further determination is made for intrinsic parameters associated with the at least one camera relative to the 3D calibration object facing in the different positions. Revised intrinsic parameters are generated from the intrinsic parameters. The revised intrinsic parameters provide a calibration output that may be applied to generate views of an item in an augmented reality application.

Abstract: Various embodiments provide methods and systems for users and business owners to share content and/or links to visual elements of a place at a physical location, and, in response to a user device pointing at a tagged place, causing the content and/or links to the visual elements of the place to be presented on the user device. In some embodiments, content and links are tied to specific objects at a place based at least in part upon one of Global Positioning System (GPS) locations, Inertial Measurement Unit (IMU) orientations, compass data, or one or more visual matching algorithms. Once the content and links are attached to the specific objects of the place, they can be discovered by a user with a portable device pointing at the specific objects in the real world.

Abstract: A method for router configuration includes at a remote router management platform, registering a first router, associated with a first router identifier, to a user account; at the first router, establishing an internet connection; at a first user electronic device, transmitting a first router configuration to the remote router management platform; and at the first router, receiving the first router configuration from the remote router management platform via the internet connection.

Abstract: Embodiments described herein are directed to allowing manipulation of visual attributes of a query image while preserving the visual attributes of a query image. A query image can be received and analyzed using a trained network to determine a set of items whose images demonstrate visual similarity to the query image across a plurality of visual attributes. Visual attributes of the query image may be manipulated to allow a user to search for items that incorporate the desired manipulated visual attributes while preserving the visual attributes of the query image. Content for at least a determined number of highest ranked, or most similar, items related to the modified visual attributes can then be provided.

Abstract: Various embodiments of systems and methods allow a distributed database to be maintained to store records of activities that may promote purchasing a product and determine whether such promotions should be rewarded. Using anonymous identifiers and a permission-based system, identities of users, products, content, content distribution systems, and marketplaces can be made available only to parties with rights to know such identities. Furthermore, systems can be put in place so that activities with a certain product, content item, etc. can be monitored without ever disclosing the identity of the product, content item, etc. In some embodiments, the database can be an immutable, append-only, distributed database such as a blockchain ledger.

Abstract: An autonomous robot with a modular conveyor belt moves materials in a warehouse or other industrial environment. Its load cells detect items on its conveyor belt and its cameras detect loading stations with stationary conveyor belts throughout the warehouse. It guides itself to a loading station using visual cues until it is close enough to load or unload items from its conveyor belt. The cart orients itself relative to the stationary conveyor belt based on the conveyor belt's direction of motion and tracks its position using wheel or visual odometry. It unloads by running the conveyor belt, which may be about 1.5 m long, at a speed of about 1 m/s for about 2 seconds. For loading, the cart detects an item on a stationary conveyor belt using a camera, positions itself next to the stationary conveyor belt, then triggers the stationary conveyor belt and its own conveyor belt to load the item.

Abstract: A computing device is used to capture image data of a physical environment. The image data is analyzed to determine color information for colors represented in the physical environment and to determine scene information that describes a room type associated with the physical environment. A palette of colors is assembled using the colors from the color information and provided for display. Upon selection of a color from the palette of colors, a product associated with the selected color and with the room type is provided for display.

Abstract: Information from a position and/or gesture detection system can be transmitted to various devices in order to enable users to interact and/or view others users. In some embodiments, video is captured that includes a current view of the body of a user. In order to prevent an unauthorized, unintended, or undesired transmission of at least part of the body image data, one or more settings or policies can be specified that can control which portions are transmitted, received, and/or displayed. For example, a user can be prompted before body image or position data is transmitted, which enables a user to control the type of data that is sent. A recipient or intermediate entity or component can also specify one or more settings or policies to control the type of data that is transmitted and/or received. In some embodiments, an external service can be utilized to manage the transmission of data.

Abstract: Various embodiments provide for visual similarity based search techniques that certain desirable visual attributes of one or more items to search for items having similar visual attributes. In order to create an electronic catalog of items that is searchable by parts-based visual attributes, the visual attributes are identified and corresponding feature vectors are extracted from the image data of each item. Thus, feature values of parts-based visual attributes of items in the electronic catalog can be determined and used to select or rank the items in response to a search query based on desirable visual attributes. To conduct a search, a user may define desirable visual attributes of one or more items. The feature vectors of the desirable visual attributes are determined and used to query the electronic catalog of items, in which items having visual attributes of similar feature vectors are selected and returned as search results.

Abstract: Various embodiments utilize a machine learning-based approach to filter items, such as apparel items, based on non-binary gender styles. For example, an electronic catalog of apparel items can be assigned gender scores on a gender scale by a neural network trained to determine a gender score of an apparel item based on an image representation of the apparel item. The neural network may be trained on training data that includes images of various apparel items with gender designations. The apparel items in the electronic catalog are assigned a gender score attribute that reflects how masculine or feminine the apparel item may be. As such, the apparel items can be organized (e.g., sorted, filtered, ranked) based on a non-binary gender score in addition to other attributes, such as item type, size, color, brand, etc. Thus, a user can include non-binary gender style as a search or filtering criteria.

Abstract: A user can capture an image of a text object of interest and have that image submitted for processing. The image can be pre-processed to improve quality and then submitted to an optical character recognition process to identify the words, characters, or strings in the image. At least some of these results can be submitted as a query to a search engine to obtain potential matches. In order to improve the accuracy of the results, information such as the titles for the results can be compared against each recognized word, character, or string from the image, including the ordering of those elements. An updated relevancy score can then be generated based on the full, ordered set. The recognized text is also analyzed to attempt to recognize model numbers or other identifiers that can be weighted more heavily as being indicative of accurate matches. Matches are selected from the re-ranked results.