Methods and systems are provided for end-to-end text recognition in digitized documents of handwritten characters over multiple lines without explicit line segmentation. An image is received. Based on the image, one or more feature maps are determined. Each of the one or more feature maps include one or more feature vectors. Based at least in part on the one or more feature maps, one or more scalar scores are determined. Based on the one or more scalar scores, one or more attention weights are determined. By applying the one or more attention weights to each of the one or more feature vectors, one or more image summary vectors are determined. Based at least in part on the one or more image summary vectors, one or more handwritten characters are determined.

Techniques for increasing robustness of a convolutional neural network based on training that uses multiple datasets and multiple tasks are described. For example, a computer system trains the convolutional neural network across multiple datasets and multiple tasks. The convolutional neural network is configured for learning features from images and accordingly generating feature vectors. By using multiple datasets and multiple tasks, the robustness of the convolutional neural network is increased. A feature vector of an image is used to apply an image-related operation to the image. For example, the image is classified, indexed, or objects in the image are tagged based on the feature vector. Because the robustness is increased, the accuracy of the generating feature vectors is also increased. Hence, the overall quality of an image service is enhanced, where the image service relies on the image-related operation.

A system for digitizing gauges, lights and other human-readable machine gauges and functions and status without interfering with the operation of the machine or requiring re-working or interfering with the existing machine wiring, signaling, electrical or mechanical elements or operating modes, or adding new digitizing equipment to the machine.

A method for reconstructing an image data set from magnetic resonance data is provided. First measurement data is captured using an image capturing device. The first measurement data is captured using temporal and/or spatial subsampling and is used for reconstructing the image data set with a compressed sensing algorithm in which a boundary condition that provided agreement with the measurement data and a target function that is used in an iterative optimization. The compressed sensing algorithm evaluates candidate data sets for the image data set are used. In the reconstruction using the compressed sensing algorithm, in addition to the first measurement data, second measurement data that is captured by a second imaging modality that is different from the first imaging modality of the first measurement data but by the same image capturing device. The second measurement data is registered to the first measurement data, by a modification of the boundary condition and/or target function.

There are included a probe that can be arranged in an imaging field of view, a horizontal drive section for causing the probe to contact a side surface of a workpiece on a stage, a display section for displaying a model image, a contact position designation section for receiving designation of contact target position information in the model image, a characteristic amount information setting section for setting characteristic amount information, a measurement setting information storage section for storing a plurality of pieces of contact target position information and the characteristic amount information, and a measurement control section for identifying a position and an attitude of the workpiece from a workpiece image by using the characteristic amount information, for identifying a plurality of contact target positions on the side surface of the workpiece where the probe should contact, based on the identified position and the identified attitude of the workpiece.

An object is recognized from image data as a target object and linked to a user based on an interaction by the user, information about the target object is obtained and a purchase of the target object is initiated.

This document describes methods, systems and computer program products directed to imaging blood cells. The subject matter described in this document can be embodied in a method of classifying white blood cells (WBCs) in a biological sample on a substrate. The method includes acquiring, by an image acquisition device, a plurality of images of a first location on the substrate, and classifying, by a processor, objects in the plurality of images into WBC classification groups. The method also includes identifying, by a processor, objects from at least some classification groups, as unclassified objects, and displaying, on a user interface, the unclassified objects and at least some of the classified objects.

Embodiments that relate to interacting with a physical object in a mixed reality environment via a head-mounted display are disclosed. In one embodiment a mixed reality interaction program identifies an object based on an image from captured by the display. An interaction context for the object is determined based on an aspect of the mixed reality environment. A profile for the physical object is queried to determine interaction modes for the object. A selected interaction mode is programmatically selected based on the interaction context. A user input directed at the object is received via the display and interpreted to correspond to a virtual action based on the selected interaction mode. The virtual action is executed with respect to a virtual object associated with the physical object to modify an appearance of the virtual object. The modified virtual object is then displayed via the display.

Apparatus, methods, and articles of manufacture for implementing crowdsourcing pipelines that generate training examples for machine learning expression classifiers. Crowdsourcing providers actively generate images with expressions, according to cues or goals. The cues or goals may be to mimic an expression or appear in a certain way, or to “break” an existing expression recognizer. The images are collected and rated by same or different crowdsourcing providers, and the images that meet a first quality criterion are then vetted by expert(s). The vetted images are then used as positive or negative examples in training machine learning expression classifiers.

In various embodiments, training objects are classified by human annotators, psychometric data characterizing the annotation of the training objects is acquired, a human-weighted loss function based at least in part on the classification data and the psychometric data is computationally derived, and one or more features of a query object are computationally classified based at least in part on the human-weighted loss function.