Google Inc.

James William Philbin

John Patrick Flynn

Hartwig Adam

Anand Pillai

Hartmut Neven

Dragomir Dimitrov Anguelov

Dmitry Kalenichenko

Gerhard Florian Schroff

Mark Cummins

Rafael Spring

Ivan Neulander

Keith Noah Snavely

Henrik C Stewenius

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating numeric embeddings of images. One of the methods includes obtaining training images; generating a plurality of triplets of training images; and training a neural network on each of the triplets to determine trained values of a plurality of parameters of the neural network, wherein training the neural network comprises, for each of the triplets: processing the anchor image in the triplet using the neural network to generate a numeric embedding of the anchor image; processing the positive image in the triplet using the neural network to generate a numeric embedding of the positive image; processing the negative image in the triplet using the neural network to generate a numeric embedding of the negative image; computing a triplet loss; and adjusting the current values of the parameters of the neural network using the triplet loss.

Generating numeric embeddings of images

A system and method of deep learning using deep networks to predict new views from existing images may generate and improve models and representations from large-scale data. This system and method of deep learning may employ a deep architecture performing new view synthesis directly from pixels, trained from large numbers of posed image sets. A system employing this type of deep network may produce pixels of an unseen view based on pixels of neighboring views, lending itself to applications in graphics generation.

Deepstereo: learning to predict new views from real world imagery

Systems and methods for a dynamic visual search engine are provided. In one example method, a criteria used to partition a set of compressed image descriptors into multiple database shards may be determined. Additionally, a size of a dynamic index may be determined. The dynamic index may represent a dynamic number of images and may be configured to accept insertion of reference images into the dynamic index that can be search against immediately. According to the method, an instruction to merge the uncompressed image descriptors of the dynamic index into the database shards of the compressed image descriptors may be received, and the uncompressed image descriptors of the dynamic index may be responsively merged into the database shards of the compressed image descriptors based on the criteria.

Systems and methods for dynamic visual search engine

A method and apparatus for enabling virtual tags is described. The method may include receiving a first digital image data and virtual tag data to be associated with a real-world object in the first digital image data, wherein the first digital image data is captured by a first mobile device, and the virtual tag data includes metadata received from a user of the first mobile device. The method may also include generating a first digital signature from the first digital image data that describes the real-world object, and in response to the generation, inserting in substantially real-time the first digital signature into a searchable index of digital images. The method may also include storing, in a tag database, the virtual tag data and an association between the virtual tag data and the first digital signature inserted into the index of digital images.

Method and apparatus for enabling virtual tags

The present application discloses systems and methods for querying an image database. In one embodiment, a method is disclosed that includes maintaining in data storage a plurality of images and a plurality of geo-feature identifiers, where each image is associated with at least one geo-feature identifier that indicates at least one geographic cell in which the image was recorded and a visual feature of the image. The method further includes receiving a query image and an indication of a requested geographic location, determining at least one target geographic cell corresponding to the requested geographic location, and identifying at least one target visual feature within the query image. The method further includes determining at least one requested geo-feature identifier based on the at least one target geographic cell and the at least one target visual feature and selecting one or more images based on the at least one requested geo-feature identifier.

Growing community of inventors

Palo Alto, CA, United States of America

James William Philbin