Abstract: A method and an apparatus for data calculation in a neural network model, and an image processing method and apparatus. The method for data calculation includes: reading weight data shared by a group of data processing of a data processing layer in a neural network model, into a GroupShared variable of a thread group of a graphics processing unit (GPU), dividing input data of the data processing layer based on the number of threads in the thread group, reading, for each group of input data, weight data corresponding to the group of input data for a data processing in the group of data processing from the GroupShared variable, and performing, by each thread in the thread group, the data processing by using a group of read input data and weight data corresponding to the group of input data, to obtain a calculation result corresponding to the group of input data.

Abstract: A non-intrusive detection method for detecting at least one pop-up window button of the pop-up window includes the following steps: retrieving a screen image on a display device; comparing the screen image with a preset screen image and generating a differential image area according the screen image and the preset screen image; determining the differential image area as the pop-up window when the differential image area is greater than an image area threshold value; selecting a plurality of contour lengths of the pop-up window matching up with a contour length threshold value by Canny edge detector; and analyzing the contour lengths according to Douglas-Peucker algorithm and an amount of endpoints to generate a contour edge corresponding to the pop-up window button.

Abstract: A first conveyor for conveying foodstuff, a second conveyor separated from the first conveyor by a gap of free space, a top camera positioned above the gap and above the first conveyor and the second conveyor, a bottom camera positioned below the gap and below the first conveyor and the second conveyor, wherein the top camera and the bottom camera capture images of the foodstuff in the gap between the first conveyor and the second conveyor for analysis.

Abstract: In an aspect for generating a 2-dimensional (2D) building mapping, a computer-implemented method may include one or more processors configured for receiving scanning data in a first file format corresponding to 3-dimensional (3D) aerial Light Detection and Ranging (LiDaR) images. Further, the method may include transforming the first file format to a second file format and transmitting the scanning data to a building semantic segmentation component. Further, the computer-implemented method may include generating semantic segmentation data corresponding to building structures defined by point clusters; generating instance segmentation data corresponding to top surfaces of the building structures; generating building boundary data for each one of the point clusters based on one or more of the semantic segmentation data and the instance segmentation data; and processing the building boundary data to generate a 2D building mapping comprising refined boundary lines for each of the point clusters.

Abstract: A method includes applying both a first dedicated functional-anatomical registration scheme to a first volume of interest to deform the first volume of interest and a second dedicated functional-anatomical registration scheme to a second volume of interest to deform the second volume of interest, wherein the first volume of interest at least partially encompasses the second volume of interest. The method includes identifying or segmenting relevant organs or anatomical structures related to a first group and a second group in the first volume of interest and the second volume of interest, respectively; generating a spatially smooth-transition weight mask that gives higher weight to image data corresponding to the identified or segmented relevant organs or anatomical structures related to the first group and the second group; and generating a final cohesive registered image volume from the first image volume and the second image volume utilizing the spatially smooth-transition weight mask.

Abstract: An image enhancement method is provided. The method includes obtaining an image, extracting high-frequency components of the image, calculating an average pixel value of pixels corresponding to the extracted high-frequency components, and performing enhancement on pixels in the image that have pixel values greater than or equal to the calculated average pixel value to obtain an enhanced image. Apparatus and non-transitory computer-readable storage medium counterpart embodiments are also provided.

Abstract: A medical image processing apparatus comprises processing circuitry configured to: receive volume data; set a symmetry plane for the volume data; generate mirrored volume data by mirroring the volume data using the symmetry plane; and generate symmetrized volume data based on the volume data and the mirrored volume data.

Abstract: A computing device is configured to detect inconsistencies on a vehicle that are typically small and difficult to detect using conventional inspection techniques. Particularly, a spotlight illuminates a target area on the vehicle. The color of the spotlight is the same color as the ambient light illuminating the vehicle but is a higher intensity. Additionally, the color of the spotlight and the ambient light is selected based on the type of inconsistencies expected to be detected. Images of the target area are then captured, digitally processed, and stored in memory, such as a database, for example. Based on the processing of these images, a weighting coefficient is computed and used to generate an enhanced contrast feature map. Inconsistencies are then detected based on the enhanced contrast feature map.

Abstract: A label processing engine receives, as inputs, raw data representative of a label and baseline data, detects a raw data object within the raw data, classifies the raw data object, and localizes the raw data object within the raw data, detects a baseline data object within the baseline data, classifies the baseline data object, and localizes the baseline data object within the baseline data. The engine recognizes corresponding text within the raw data object and the baseline data object and extracts the corresponding text within the raw data object and the baseline data object, reassembles the corresponding text of the raw data object and the baseline data object into respective lines of text, compares the respective lines of text with one another, and issues a notification based on the comparison.

Abstract: Video frames from a video are compressed into a single image or a single data structure that represents a unique visual flowprint or visual signature for a given activity being modeled from the video frames. The flowprint comprises a computed summary of the original pixel values associated with the video frames within the single image and the flowprint is specific to movements occurring within the video frames that are associated with the given activity. In an embodiment, the flowprint is provided as input to a machine-learning algorithm to allow the algorithm to perform object tracking and monitoring from the flowprint rather than from the video frames of the video, which substantially improves processor load and memory utilization on a device that executes the algorithm, and substantially improved responsiveness of the algorithm.

Abstract: Embodiments are disclosed for translating an image from a source visual domain to a target visual domain. In particular, in one or more embodiments, the disclosed systems and methods comprise a training process that includes receiving a training input including a pair of keyframes and an unpaired image. The pair of keyframes represent a visual translation from a first version of an image in a source visual domain to a second version of the image in a target visual domain. The one or more embodiments further include sending the pair of keyframes and the unpaired image to an image translation network to generate a first training image and a second training image. The one or more embodiments further include training the image translation network to translate images from the source visual domain to the target visual domain based on a calculated loss using the first and second training images.

Abstract: Methods and apparatus for blending unknown pixels in overlapping images. In one embodiment, an action camera captures two hyper-hemispherical fisheye images that are stitched to a 360° panorama. In order to remove exposure differences between the two cameras, the images are pre-processed prior to multiband blending. The pre-processing leverages image information from pixels to make informed guesses about pixels that were not captured. In particular, various pixels with different knowability (e.g., known, unknown, consistent, and conflicting) may be handled differently so as to emphasize/de-emphasize their importance in pre-processing.

Abstract: A method includes obtaining at least one 2-D image dataset of semiconductor structures formed on a wafer including one or more defects during a wafer run of a wafer using a predefined fabrication process. The method also includes determining, based on at least one machine-learning algorithm trained on prior knowledge of the fabrication process and based on the at least one 2-D image dataset, one or more process deviations of the wafer run from the predefined fabrication process as a root cause of the one or more defects. A 3-D image dataset may be determined as a hidden variable.

Abstract: Disclosed is a method and system for achieving optimal separable convolutions, the method is applied to image analyzing and processing and comprises steps of: inputting an image to be analyzed and processed; calculating three sets of parameters of a separable convolution: an internal number of groups, a channel size and a kernel size of each separated convolution, and achieving optimal separable convolution process; and performing deep neural network image process. The method and system in the present disclosure adopts implementation of separable convolution which efficiently reduces a computational complexity of deep neural network process. Comparing to the FFT and low rank approximation approaches, the method and system disclosed in the present disclosure is efficient for both small and large kernel sizes and shall not require a pre-trained model to operate on and can be deployed to applications where resources are highly constrained.

Abstract: A method for identifying an individual of an oplegnathus punctatus based on a convolutional neural network is provided. Target initial positioning involves three continuous convolutional layers and an average pooling layer. A region of feature interest point is obtained, a hyperparameter candidate box is set to obtain a region, thereby obtaining an approximate position of a target object. II_Net backbone convolutional neural network includes six convolutional layers and four pooling layers, which includes a LeakyReLU activation function used as an activation function of the first convolutional layer, convolutional network layers of Alexnet and parameter data, and a maximum pooling layer of an overlapped pooling structure. Fully connected layers use a genetic algorithm to improve data transmission between layers. With the established model, identification of the individual of the oplegnathus punctatus is performed by using test data.

Abstract: A method, system and mobile device can be used by a subject in diagnosing a disease or virus or other illness. The system can capture and analyze olfactory information providing a diagnosis based on the olfactory information. The system can also capture and output biometric data corresponding to the subject and can further include at least one camera or video sensor that can result in a diagnosis or a microphone or other acoustic sensor that can result in another diagnosis. Other sensors providing other corresponding diagnoses can be included. A sensor fusion component can receive and combine the biometric data and the various diagnoses results and further determine a confidence score. An event record creator compiles the biometric data and the confidence scores to create an event record having a higher confidence score with respect to a final diagnosis result. A data storage device stores the event record.

Abstract: The invention addresses the problem of improving labor of a registration operation of a product and enabling a checkout operator to recognize a recognition result. In order to solve the problem, the invention provides a registration apparatus (10) including: an image acquisition unit (11) that acquires an image obtained by imaging a placement surface of a table, on which a product is placed; an analysis unit (12) that recognizes the product included in the image, a registration unit (14) that registers the recognized product as a checkout target, and an output unit (13) that outputs a name of the recognized product by voice.

Abstract: An oral-area positioning device is provided in the invention. The oral-area positioning device includes a storage circuit, a positioning circuit and a calculation circuit. The storage circuit stores information corresponding to a plurality of oral areas. The positioning circuit obtains a target image from an oral-image extracting device, and obtains a first position estimation result according to the information corresponding to the plurality of oral areas and a first algorithm. The positioning device obtains a second position estimation result at least according to the information corresponding to the plurality of oral areas, a second algorithm and a reference image position of a reference image, wherein the reference image position is one of the oral areas. The calculation circuit generates a third position estimation result according to the first position estimation result and the second position estimation result.

Abstract: A system and method for augmenting an original OCT (optical coherence tomography) image includes a controller having a processor and a tangible, non-transitory memory on which instructions are recorded. The system includes one or more learning modules selectively executable by the controller. The learning modules are trained by a training network with a training dataset having a plurality of training ultrasound bio-microscopy images and respective training OCT images. Execution of the instructions by the processor causes the controller to obtain the original OCT image, captured through an OCT device. The controller is configured to generate an augmented OCT image based in part on the original OCT image, by executing the (trained) learning modules. The augmented OCT image at least partially extends a peripheral portion of the original OCT image.

Abstract: Techniques performed by a data processing system include establishing an online presentation session for conducting an online presentation, receiving first media streams comprising presentation content from the first computing device, receiving second media streams from the second computing devices of a subset of the plurality of participants, the second media streams including audio content, video content, or both of the subset of the plurality of participants, analyzing the first media streams using first machine learning models to generate feedback results, analyzing the set of second media streams to identify first reactions by the participants to obtain reaction information, automatically analyzing the feedback results and the reactions to identify discrepancies between the feedback results and the reactions, and automatically updating one or more parameters of the machine learning models based on the discrepancies to improve the suggestions for improving the online presentation.