Abstract: Methods, systems and computer program products are provided for prefetching information and/or (pre)allocating computing resources based on predicting classification labels with temporal data. A trained temporal classification model forecasts events (e.g., too numerous for individual modeling) by predicting classification labels indicating whether events will occur, or a number of occurrences of the events, during each of a plurality of future time intervals. Time-series datasets, indicating whether events occurred, or a number of occurrences of the events, during each of a plurality of past time intervals, are transformed into temporal classification datasets. Classifications may be based, at least in part, on extracted features, such as data seasonality, temporal representation, statistical and/or real-time features. Classification labels are used to determine whether to take one or more actions, such as, for example, prefetching information or (pre)allocating a computing resource.

Abstract: A system and method leverages understanding of complex dribbling video clips by representing a video sequence with a single Dribble Energy Image (DEI) that is informative for dribbling styles recognition. To overcome the shortage of labelled data, a dataset of soccer video clips employs Mask-RCNN to segment out dribbling players and OpenPose to obtain joints information of dribbling players. To solve issues caused by camera motions in highlight soccer videos, the system registers a video sequence to generate a single image representation DEI and dribbling styles classification.

Abstract: An electronic device having a neural network framework for estimating optical flow is provided. The electronic device is connected to an image acquiring unit, which acquires images to be analyzed. The electronic device includes a storage unit, a feature extraction unit, an optical flow estimation unit and a refining unit. The storage unit stores a feature extraction module. The feature extraction unit is connected to the image acquiring unit and the storage unit. The optical flow estimation unit is connected to the feature extraction unit to generate an estimated optical flow. The refining unit is connected to the optical flow estimation unit to input the estimated optical flow to a refining module to obtain an estimated optical flow result. A method for estimating optical flow is also provided to reduce the number of training parameters required for estimating optical flow, thereby reducing a training time and improving training stability.

Abstract: An information processing device includes a processor. The processor obtains an input image, inputs the input image to a machine learning model that executes classification likelihood calculation processing to obtain, for each of candidate objects in the input image, likelihoods of belonging to the plurality of classes, executes first determination on whether or not each of the candidate objects is classified as a first class of the plurality of classes using a likelihood of belonging to the first class that is a likelihood having a negative correlation with likelihoods of belonging to other classes, executes second determination on whether or not each of the candidate objects that have been determined in the first determination as a non-first class is classified as the other classes, and outputting a result of classifying the candidate objects included in the input image using a result of the second determination.

Abstract: Devices and techniques are generally described for estimating three-dimensional pose data. In some examples, a first machine learning network may generate first three-dimensional (3D) data representing input 2D data. In various examples, a first 2D projection of the first 3D data may be generated. A determination may be made that the first 2D projection conforms to a distribution of natural 2D data. A second machine learning network may generate parameters of a 3D model based at least in part on the input 2D data and based at least in part on the first 3D data. In some examples, second 3D data may be generated using the parameters of the 3D model.

Abstract: An electronic device having a biometric sensor is provided. The electronic device includes a first region including a plurality of first pixels arranged in a first manner and a second region including a plurality of second pixels arranged in a second manner, a biometric sensor disposed in at least a part of the first region, and a processor electrically coupled with the display and the biometric sensor, and configured to receive a user input through the first region, and control the biometric sensor to detect biometric information corresponding to the user input.

Abstract: Systems and methods to localize objects for mapping applications may comprise a vehicle having an imaging device, a location sensor, and an edge processor. Using imaging data from the imaging device, location data from the location sensor, and bounding box data associated with objects, three-dimensional models of environments may be reconstructed using structure from motion algorithms and/or direct triangulation algorithms. After aligning the reconstructions to real-world environments based on the location data, objects may be accurately localized relative to real-world environments.

Abstract: An object tracking device includes an image buffer configured to store a plurality of images included in a video, a detection part configured to detect an object position by executing a detection process with respect to an object reflected in the plurality of images, a tracking part configured to execute a tracking process with respect to a tracking-image string representing a scope of images ranging from an image producing an object-detection result and an image producing a next object-detection process, and an integration part configured to calculate an integrative object position by integrating a detection result and a tracking result. Herein, the scope of images is determined based on an execution timing or a termination timing of the detection process, thus executing the detection process and the tracking process in parallel.

Abstract: A field-of-view at a first modeled target location of a first target disposed on a specimen can be configured, which can include moving the stage relative to the detector. The first modeled target location is determined by summing a first design target location and a navigational error provided by an online model. A first image of the field-of-view is grabbed using the detector. The field-of-view at a second modeled target location of a second target disposed on the specimen is configured. Concurrent with configuring the field-of-view at the second modeled target location, using a processor, the position of a first actual target location is determined using the first image. The online model is updated with a difference between the first design target location and the first actual target location.

Abstract: An electronic device is provided. The electronic device includes: a display panel, a display controller, and a host. The display controller is configured to control displaying of the display panel. The host is electrically connected to the display controller. The host executes an application to render a display image, and transmits the display image to the display controller through an image-transmission channel between the host and the display controller. The display controller detects whether pixels in a specific area of the display image satisfy a predetermined condition. In response to the pixels in the specific area satisfying the predetermined condition, the display controller triggers a virtual input signal, and transmits the virtual input signal to the host through a data-transmission channel between the host and the display controller, so that the host executes a specific operation corresponding to the virtual input signal.

Abstract: Aspects of the present disclosure describe systems, methods and structures for an efficient multi-person posetracking method that advantageously achieves state-of-the-art performance on PoseTrack datasets by only using keypoint information in a tracking step without optical flow or convolution routines. As a consequence, our method has fewer parameters and FLOPs and achieves faster FPS. Our method benefits from our parameter-free tracking method that outperforms commonly used bounding box propagation in top-down methods. Finally, we disclose tokenization and embedding multi-person pose keypoint information in the transformer architecture that can be re-used for other pose tasks such as pose-based action recognition.

Abstract: The present disclosure provides an image processing method, device and apparatus, and a computer-readable storage medium. The image processing method includes: obtaining an image and detecting multiple targets in the image; obtaining categories of multiple detected targets and first parameters of a detection frame corresponding to each detected target, and ranking all detected targets according to the first parameters in a descending order to obtain a dataset; obtaining a target with a highest first parameter in the dataset as a first detected target, and detecting whether the first parameter of a first detection frame corresponding to the first detected target satisfies a preset category condition of a category to which the first detected target belongs; detecting whether the first detection frame corresponding to the first detected target overlaps with a detection frame corresponding to a respective one of the plurality of detected targets except the first detected target.

Abstract: An image matching device that performs matching between a first image and a second image includes: a frequency characteristic acquisition unit configured to acquire a frequency characteristic of the first image and a frequency characteristic of the second image; a frequency characteristic synthesizing unit configured to synthesize the frequency characteristic of the first image and the frequency characteristic of the second image to generate a synthesized frequency characteristic; a determination unit configured to perform frequency transformation on the synthesized frequency characteristic to calculate a correlation coefficient map whose resolution coincides with a target resolution, and perform matching between the first image and the second image based on a matching score calculated from the correlation coefficient map; and a regulation unit configured to regulate the target resolution based on the matching score.

Abstract: Systems and methods for road typology scene annotation are provided. A method for road typology scene annotation includes receiving an image having a road scene. The image is received from an imaging device. The method populates, using a machine learning model, a set of attribute settings with values representing the road scene. An annotation interface is implemented and configured to adjust values of the attribute settings to correspond with the road scene. Based on the values of the attribute settings, a simulated overhead view of the respective road scene is generated.

Abstract: The present disclosure provides an image inpainting method, an image inpainting system, and a flat panel detector. The method includes: acquiring a first image, performing connected domain analysis on the first image, and extracting the defective pixels in the first image; and classifying the defective pixels into isolated point defects, cluster defects, and bad line defects according to the size and shape of the connected domain; outputting a second image and inpainting the second image according to the type of the defective pixels. This method can not only effectively inpaint defective pixels, but also well preserve the details of the original image.

Abstract: Provided are a matching method and apparatus, an electronic device, and a computer-readable storage medium. The method includes: obtaining a to-be-matched image; extracting at least one to-be-matched feature point from the to-be-matched image; for one of at least one reference image, performing matching between the at least one to-be-matched feature point and at least one reference feature point extracted from the reference image to obtain a distance set corresponding to the reference image, the distance set corresponding to the reference image being a distance set composed of distances between the at least one to-be-matched feature point and the at least one reference feature point, the one reference image including an image of a standard sample, and different reference images including different standard samples; and determining, based on the distance set corresponding to each reference image, whether the to-be-matched image matches each reference image.

Abstract: Systems, methods, tangible non-transitory computer-readable media, and devices associated with object localization and generation of compressed feature representations are provided. For example, a computing system can access training data including a source feature representation and a target feature representation. An encoded target feature representation can be generated based on the target feature representation and a machine-learned encoding model. A binarized target feature representation can be generated based on the encoded target feature representation and lossless binarization operations. A reconstructed target feature representation can be generated based on the binarized target feature representation and a machine-learned decoding model. A matching score for the source feature representation and the reconstructed target feature representation can be determined. A loss associated with the matching score can be determined.

Abstract: Embodiments of the present invention are generally directed to generating figure captions for electronic figures, generating a training dataset to train a set of neural networks for generating figure captions, and training a set of neural networks employable to generate figure captions. A set of neural networks is trained with a training dataset having electronic figures and corresponding captions. Sequence-level training with reinforced learning techniques are employed to train the set of neural networks configured in an encoder-decoder with attention configuration. Provided with an electronic figure, the set of neural networks can encode the electronic figure based on various aspects detected from the electronic figure, resulting in the generation of associated label map(s), feature map(s), and relation map(s).

Abstract: A curved display and a method for binding a cover glass of the curved display are provided. The curved display includes a display module, a frame body and a cover glass. The frame body has a first flat surface and a second flat surface opposite to the first flat surface, in which the first flat surface is adhered to the display module. The cover glass has a binding flat surface and an application surface opposite to the binding flat surface, in which the binding flat surface is adhered to the second flat surface of the frame body, and the application surface is a surface with curvature. The second flat surface of the frame body is set with a first alignment mark, and the binding flat surface of the cover glass is set with a second alignment mark, and the first alignment mark corresponds to the second alignment mark.

Abstract: Methods and systems for diagnosing a semiconductor wafer are provided. A plurality of raw images of the semiconductor wafer are obtained according to GDS information regarding a layout of a target die, by an inspection apparatus. A first image-based comparison is performed on the raw images, so as to provide a comparison result, by a determining circuitry. The comparison result indicates whether an image difference is present between the images. One of the raw images having the image difference is assigned as a defect image. A second image-based comparison is performed on a reference image and the defect image, so as to classify a defect type of the image difference, by the determining circuitry. The layout of the target die includes a circuit with a duplicate layout formed by a plurality of same cells. The number of the plurality of raw images is greater than 2.