Google Inc.

Verily Life Sciences LLC

Deepmind Technologies Limited

Dilip Krishnan

Forrester H Cole

William T Freeman

Ce Liu

Nathan Silberman

Dumitru Erhan

David Benjamin Belanger

Konstantinos Bousmalis

Aaron Joseph Maschinot

Aaron Yehuda Sarna

Piotr Teterwak

Tiantian Zha

Chen Wang

Inbar Mosseri

Sven Adrian Gowal

Mark Desnoyer

Krishnamurthy Dvijotham

Chongli Qin

Robert Stanforth

James Martens

Alhussein Fawzi

Prannay Khosla

Aaron Sarna

Josh Livni

Soham De

Yaniv Ovadia

George Trigeorgis

Yonglong Tian

Peter Massaro

Jingbin Wang

Omer Gimenez Llach

Mikael Bonnevie

Victor Criswell

Nigel Snoad

Michael Krainin

Huiwen Chang

Mengjie Wang

Yuanzhen Li

Hossein Talebi

David Martin Dohan

Phillip John Isola

Xiao Feng

Yihui Wang

Han Xu

Aj Maschinot

Yi Han

Raanan Sayag

Haifeng Gong

Wenchao Tong

Shuchao Bi

Philip John Isola

Daniel Newburger

Michael Spencer Krainin

Mikael Pierre Bonnevie

The present disclosure provides an improved training methodology that enables supervised contrastive learning to be simultaneously performed across multiple positive and negative training examples. In particular, example aspects of the present disclosure are directed to an improved, supervised version of the batch contrastive loss, which has been shown to be very effective at learning powerful representations in the self-supervised setting. Thus, the proposed techniques adapt contrastive learning to the fully supervised setting and also enable learning to occur simultaneously across multiple positive examples.

Supervised contrastive learning with multiple positive examples

To adjust an aspect ratio of an image to match the aspect ratio of a display area for presenting the image, a computing device receives an image having a first aspect ratio, and obtains a second aspect ratio for a display area of a display in which to present the image, where the second aspect ratio is different from the first aspect ratio. The computing device extends the image to include one or more additional features which were not included in the image. Additionally, the computing device automatically crops the extended image around an identified region of interest by selecting a portion of the extended image that has an aspect ratio which matches the second aspect ratio of the display area, and provides the cropped image for presentation within the display area of the display.

Flexible image aspect ratio using machine learning

The present disclosure provides systems and methods that perform face reconstruction based on an image of a face. In particular, one example system of the present disclosure combines a machine-learned image recognition model with a face modeler that uses a morphable model of a human's facial appearance. The image recognition model can be a deep learning model that generates an embedding in response to receipt of an image (e.g., an uncontrolled image of a face). The example system can further include a small, lightweight, translation model structurally positioned between the image recognition model and the face modeler. The translation model can be a machine-learned model that is trained to receive the embedding generated by the image recognition model and, in response, output a plurality of facial modeling parameter values usable by the face modeler to generate a model of the face.

Face reconstruction from a learned embedding

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating realistic extensions of images. In one aspect, a method comprises providing an input that comprises a provided image to a generative neural network having a plurality of generative neural network parameters. The generative neural network processes the input in accordance with trained values of the plurality of generative neural network parameters to generate an extended image. The extended image has (i) more rows, more columns, or both than the provided image, and (ii) is predicted to be a realistic extension of the provided image. The generative neural network is trained using an adversarial loss objective function.

Image extension neural networks

Insects can be classified into a category (e.g., sex category, species category, size category, etc.) using a variety of different classification approaches including, for example, an industrial vision classifier and/or a machine learning classifier. At least some classification approaches may be used in real-time to make real-time decisions and others can be used to validate earlier-made real-time decisions.

Predictive classification of insects

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network. One of the methods includes processing each training input using the neural network and in accordance with the current values of the network parameters to generate a network output for the training input; computing a respective loss for each of the training inputs by evaluating a loss function; identifying, from a plurality of possible perturbations, a maximally non-linear perturbation; and determining an update to the current values of the parameters of the neural network by performing an iteration of a neural network training procedure to decrease the respective losses for the training inputs and to decrease the non-linearity of the loss function for the identified maximally non-linear perturbation.

Training more secure neural networks by using local linearity regularization

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images using an image processing neural network system. One of the system includes a shared encoder neural network implemented by one or more computers, wherein the shared encoder neural network is configured to: receive an input image from a target domain; and process the input image to generate a shared feature representation of features of the input image that are shared between images from the target domain and images from a source domain different from the target domain; and a classifier neural network implemented by the one or more computers, wherein the classifier neural network is configured to: receive the shared feature representation; and process the shared feature representation to generate a network output for the input image that characterizes the input image.

Domain separation neural networks

Insects can be localized and classified using a predictive model. To begin, image data is obtained that corresponds to the insects. Using a predictive model, samples of the image data are evaluated to determine whether the image portions include an insect and, if so, into what category the insect should be classified (e.g., male/female, species A/species B, etc.).

Predictive models for visually classifying insects

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images using an image processing neural network system. One of the systems includes a domain transformation neural network implemented by one or more computers, wherein the domain transformation neural network is configured to: receive an input image from a source domain; and process a network input comprising the input image from the source domain to generate a transformed image that is a transformation of the input image from the source domain to a target domain that is different from the source domain.

Transforming source domain images into target domain images

A system and method for generating cartoon images from photos are described. The method includes receiving an image of a user, determining a template for a cartoon avatar, determining an attribute needed for the template, processing the image with a classifier trained for classifying the attribute included in the image, determining a label generated by the classifier for the attribute, determining a cartoon asset for the attribute based on the label, and rendering the cartoon avatar personifying the user using the cartoon asset.

Generating cartoon images from photos

Growing community of inventors

Arlington, MA, United States of America

Dilip Krishnan