Google Inc.

Boris Bluntschli

Daniel Cotting

Daniel Raynaud

Harshit Kharbanda

Louis Wang

Vibhuti Mahajan

Wojciech Stanislaw Smietanka

Roland Peter Kehl

Krzysztof Sikora

Martin Stefcek

Nicolas Dumazet

Systems and methods for image understanding can include one or more object recognition systems and one or more vision language models to generate an augmented language output that can be both scene-aware and object-aware. The systems and methods can process an input image with an object recognition model to generate an object recognition output descriptive of identification details for an object depicted in the input image. The systems and methods can include processing the input image with a vision language model to generate a language output descriptive of a predicted scene description. The object recognition output can then be utilized to augment the language output to generate an augmented language output that includes the scene understanding of the language output with the specificity of the object recognition output.

Instance level scene recognition with a vision language model

A system and computer implemented method for detecting the location of a mobile device using semantic indicators is provided. The method includes receiving, using one or more processors, a plurality of images captured by a mobile device at an area. The area is associated with a set of candidate locations. Using the one or more processors, one or more feature indicators associated with the plurality of images are detected. These feature indicators include semantic features related to the area. The semantic features are compared with a plurality of stored location features for the set of candidate locations. In accordance with the comparison, a location from the set of candidate locations is selected to identify an estimated position of the mobile device.

Detecting the location of a mobile device based on semantic indicators

Systems and methods for determining a geocode for an image based on user-provided search queries and corresponding user selections are provided. One example method includes determining a selection value for each of a plurality of search strings associated with an image based at least in part on user selection data. The method includes generating a textual document for the image based at least in part on the selection values. The textual document includes one or more of the plurality of search strings. The method includes identifying a plurality of geographic entities by analyzing the textual document using a textual processor. The method includes selecting one of the plurality of geographic entities as a primary geographic entity and associating, by the one or more computing devices, a geocode associated with the primary geographic entity with the image.

Geocoding of images based on search queries

Aspects of the disclosure relate generally to systems and methods for geographically locating images. For example, images from different sources may be associated with different types of location information or simply none at all. In order to reduce inconsistency among images, location information may be gathered for an image using bitmap processing, metadata processing, and information retrieved from where the image was found. This location information can be filtered to remove less reliable or conflicting information. Images may be clustered based on their appearance on an interactive online resource that corresponds to a user-defined event, based on image similarity, and by their appearance in a user photo album. The location information of the images of a cluster is then copied to all of the other images of that cluster.

Geographically locating and posing images in a large-scale image repository and processing framework

Systems and methods for the processing of images are provided. In particular, a candidate image can be obtained for processing. The candidate image can have one or more associated image categorization parameters. One or more pixel groups can then be detected in the candidate image and the one or more pixel groups can be associated with semantic data. At least one reference image can then be identified based at least in part on the semantic data of the one or more pixel groups. Once the at least one reference image has been identified, a plurality of adjustment parameters can be determined. One or more pixel groups from the candidate image can then be processed to generate a processed image based at least in part on the plurality of adjustment parameters.

Growing community of inventors

Zurich, Switzerland

Boris Bluntschli