A recognition processor calculates a recognition score indicating a possibility at which a predetermined target object is included in a partial region of a captured picture. In a case where a picture size of the partial region is smaller than a threshold value, the recognition processor calculates a recognition score using first recognition dictionary data generated by machine learning that sets a picture having a picture size smaller than a predetermined value as an input picture. In a case where a picture size of the partial region is larger than or equal to the threshold value, the recognition processor calculates a recognition score using second recognition dictionary data generated by machine learning that sets a picture having a picture size larger than or equal to the predetermined value as an input picture.

An editing device acquires a first image in which an occupant of a vehicle has been imaged in association with a time point in a time series and a second image in which scenery around the vehicle has been imaged in association with a time point in a time series, acquires first index information indicating feelings of the occupant when the first image has been captured on the basis of the first image, and extracts the first image and the second image from first images of the time series and second images of the time series on the basis of the first index information and the time point associated with the first image based on the first index information to generate a library including the extracted images.

Methods and systems for performing machine learning include selecting first and second training data from one or more training sets in one or more databases. Mixed training data is formed by subtracting a value of each data element in the second training data from a value of a corresponding data element in the first training data. A machine learning process is trained using the mixed training to augment data used by the machine learning process.

A mechanism for creating/synthesizing realistic training data, for training a machine-learning model, using anatomical knowledge. An anatomical model can be obtained. Information from annotated training data entries (i.e. “ground truth” information), can be used to model the anatomical variation, from the obtained model, in the population of the training data. This anatomical model can then be modified, e.g. incorporating some random factors, in order to generate one or more augmented models of realistic anatomies. The augmented anatomy is then transferred from the model domain to the data entry domain to thereby generate a new data entry or data entries for training a machine-learning model. This latter process can be achieved in various ways, e.g. using GANs, such as CycleGANs and label images, or deformation vector fields.

A system for capturing images for training an item identification model obtains an identifier of an item. The system detects a triggering event at a platform, where the triggering event corresponds to a user placing the item on a platform. The system causes the platform to rotate. The system causes at least one camera to capture an image of the item while the platform is rotating. The system extracts a set of features associated with the item from the image. The system associates the item to the identifier and the set of features. The system adds a new entry to a training dataset of the item identification model, where the new entry represents the item labeled with the identifier and the set of features.

In one aspect, there is provided a method performed by one or more data processing apparatus, the method including receiving a training dataset having multiple training examples, where each training example includes: (i) an image, and (ii) a segmentation defining a target region of the image that has been classified as including pixels in a target category. The method further includes determining a respective refined segmentation for each training example, including, for each training example, processing the target region of the image defined by the segmentation for the training example using a de-noising neural network to generate a network output that defines the refined segmentation for the training example. The method further includes training a segmentation machine learning model on the training examples of the training dataset, including, for each training example training the segmentation machine learning model to process the image included in the training example to generate a model output that matches the refined segmentation for the training example.

A method of adaptively updating an enrollment database is disclosed. The method may include extracting a first feature vector from an input image, the input image including a face of a user, determining whether to enroll the input image in the enrollment database based on the first feature vector, second feature vectors of enrollment images and a representative vector, the second feature vectors of the enrollment images being enrolled in the enrollment database, and the representative vector representing the second feature vectors, and enrolling the input image in the enrollment database based on a result of the determining.

A method, apparatus, computer system, and computer program product for managing a device. The method detects, by a computer system, a physical handling of the device to form a physical handling pattern for the device. The method determines, by the computer system, a baseline physical handling pattern for the device, wherein the baseline physical handling pattern for the device meets a set of handling metrics for the device. The method initiates, by the computer system, a set of actions in response to the physical handling pattern for the device deviating from the baseline physical handling pattern for the device.

Methods, systems, and computer readable storage media for using image processing to develop a library of facial expressions. The system can receive digital video of at least one speaker, then execute image processing on the video to identify landmarks within facial features of the speaker. The system can also identify vectors based on the landmarks, then assign each vector to an expression, resulting in a plurality of speaker expressions. The system then scores the expressions based on similarity to one another, and creates subsets based on the similarity scores.

A data generating method includes obtaining first sample data, determining a type of the first sample data and a corresponding data expansion method, expanding the first sample data according to the determined data expansion method to generate second sample data, and dividing the first sample data and the second sample data into a training set and a verification set according to a preset rule. A data model is trained according to the training set, and the data model is verified according to the verification set after training.