According to an example aspect of the present invention, there is provided an apparatus comprising memory configured to store convolutional artificial neural network information comprising at least one filter definition, and at least one processing core configured to generate, from a preceding layer, a convolutional result of a succeeding layer of the artificial neural network in accordance with the at least one filter definition, and generate, from the convolutional result, an activation result of the succeeding layer by using an activation function, the activation function taking three arguments, the three arguments being derived from the convolutional result.

Briefly, embodiments of methods and/or systems of generating preference indices for contiguous portions of digital images are disclosed. For one embodiment, as an example, parameters of a neural network may be developed to generate object labels for digital images. The developed parameters may be transferred to a neural network utilized to generate signal sample value levels corresponding to preference indices for contiguous portions of digital images.

A method for implementing source-free domain adaptive detection is presented. The method includes, in a pretraining phase, applying strong data augmentation to labeled source images to produce perturbed labeled source images and training an object detection model by using the perturbed labeled source images to generate a source-only model. The method further includes, in an adaptation phase, training a self-trained mean teacher model by generating a weakly augmented image and multiple strongly augmented images from unlabeled target images, generating a plurality of region proposals from the weakly augmented image, selecting a region proposal from the plurality of region proposals as a pseudo ground truth, detecting, by the self-trained mean teacher model, object boxes and selecting pseudo ground truth boxes by employing a confidence constraint and a consistency constraint, and training a student model by using one of the multiple strongly augmented images jointly with an object detection loss.

An information processing apparatus acquires video data that includes target objects including a person and an object, and identifies a relationship between the target objects in the acquired video data, by inputting the acquired video data to a first machine learning model. The information processing apparatus identifies a behavior of the person in the video data by using a feature value of the person included in the acquired video data. The information processing apparatus predicts one of a future behavior and a future state of the person by comparing the identified behavior of the person and the identified relationship with a behavior prediction rule that is set in advance.

Various embodiments described herein relate to techniques for forecasting with state transitions and confidence factors. In this regard, a system is configured to segment data associated with one or more assets to determine a set of classifications for one or more attributes related to the one or more assets. The system is also configured to generate a state machine associated with a Markov chain model based on the set of classifications for the data. Furthermore, the system is configured to perform a machine learning process associated with the state machine to determine one or more behavior changes associated with the one or more attributes related to the one or more assets. The system is also configured to predict, based on the one or more behavior changes associated with the one or more attributes related to the one or more assets, a change in demand data for the one or more assets during a future interval of time.

A method for detecting boundaries of lanes on a road is presented. The method comprises receiving, by one or more processors from an imaging system, a set of pixels associated with lane markings. The method further includes partitioning, by the one or more processors, the set of pixels into a plurality of groups. Each of the plurality of groups is associated with one or more control points. The method further includes generating, by the one or more processors, a spline that traverses the control points of the plurality of groups. The spline traversing the control points describes a boundary of a lane.

Systems and methods are described herein for processing electronic medical images to optimize a review order of pathology cases. For example, a plurality of variables and one or more constraints may be received along with a plurality of pathology cases. Each case of the plurality of pathology cases may include one or more medical images of at least one pathology specimen associated with a patient. The medical images from each case, the plurality of variables, and the one or more constraints may be provided as input to a trained system. A sequential order for user review of the plurality of cases to optimize one or more of the plurality of variables based on the one or more constraints may be received as output of the trained system. Each case of the plurality of cases may be automatically provided to a user for review according to the sequential order.

The present disclosure relates to an object selection system that automatically detects and selects objects in a digital image utilizing a large-scale object detector. For instance, in response to receiving a request to automatically select a query object with an unknown object class in a digital image, the object selection system can utilize a large-scale object detector to detect potential objects in the image, filter out one or more potential objects, and label the remaining potential objects in the image to detect the query object. In some implementations, the large-scale object detector utilizes a region proposal model, a concept mask model, and an auto tagging model to automatically detect objects in the digital image.

A method is provided for Cross Video Temporal Difference (CVTD) learning. The method adapts a source domain video to a target domain video using a CVTD loss. The source domain video is annotated, and the target domain video is unannotated. The CVTD loss is computed by quantizing clips derived from the source and target domain videos by dividing the source domain video into source domain clips and the target domain video into target domain clips. The CVTD loss is further computed by sampling two clips from each of the source domain clips and the target domain clips to obtain four sampled clips including a first source domain clip, a second source domain clip, a first target domain clip, and a second target domain clip. The CVTD loss is computed as |(second source domain clip−first source domain clip)−(second target domain clip−first target domain clip)|.

A method for performing automatic visual inspection includes: capturing visual information of an object using a scanning system including a plurality of cameras; extracting, by a computing system including a processor and memory, one or more feature maps from the visual information using one or more feature extractors; classifying, by the computing system, the object by supplying the one or more feature maps to a complex classifier to compute a classification of the object, the complex classifier including: a plurality of simple classifiers, each simple classifier of the plurality of simple classifiers being configured to compute outputs representing a characteristic of the object; and one or more logical operators configured to combine the outputs of the simple classifiers to compute the classification of the object; and outputting, by the computing system, the classification of the object as a result of the automatic visual inspection.