Abstract: Embodiments are disclosed for training a neural network classifier to learn to more closely align an input image with its attribution map. In particular, in one or more embodiments, the disclosed systems and methods comprise receiving a training image comprising a representation of one or more objects, the training image associated with at least one label for the representation of the one or more objects, generating a perturbed training image based on the training image using a neural network, and training the neural network using the perturbed training image by minimizing a combination of classification loss and attribution loss to learn to align an image with its corresponding attribution map.

Abstract: A method includes obtaining, using at least one processor, an input image frame. The method also includes identifying, using the at least one processor, one or more regions of the input image frame containing redundant information. In addition, the method includes performing, using the at least one processor, an image processing task using the input image frame. The image processing task is guided based on the one or more identified regions of the input image frame. The method may further include obtaining, using the at least one processor, a coarse depth map associated with the input image frame. Performing the image processing task may include refining the coarse depth map to produce a refined depth map, where the refining of the coarse depth map is guided based on the one or more identified regions of the input image frame.

Abstract: A vision-LiDAR fusion method and system based on deep canonical correlation analysis are provided. The method comprises: collecting RGB images and point cloud data of a road surface synchronously; extracting features of the RGB images to obtain RGB features; performing coordinate system conversion and rasterization on the point cloud data in turn, and then extracting features to obtain point cloud features; inputting point cloud features and RGB features into a pre-established and well-trained fusion model at the same time, to output feature-enhanced fused point cloud features, wherein the fusion model fuses RGB features to point cloud features by using correlation analysis and in combination with a deep neural network; and inputting the fused point cloud features into a pre-established object detection network to achieve object detection. A similarity calculation matrix is utilized to fuse two different modal features.

Abstract: Systems and methods that facilitate determining interaction between medications and the brain using a brain measure and a brain model. Hidden nervous system states are difficult to predict, diagnose, and treat with therapeutic medications. A Dual Neural Machine Translation (d-NMT) algorithmic system that utilizes sets of parameters for a relapsing-remitting MS model based on patient medical records and adjusts a method of parameterization to produce a model that can match patients' medical records and medical images. These parameters are can be used by a therapeutic determining model to recommend therapies, doses, and time courses accurately.

Abstract: Disclosed are computer systems and computer-implemented methods for rapid diagnostic test result interpretation platform employing computer vision. The system can visually analyze the test device in real-time by receiving image streams of the test device from the camera system. The detection system can be used to identify the test result in the test region. In some embodiments, the test result is binary (e.g., Positive/Negative). In some embodiments, the test result is in semi-quantitative formats (e.g., 1-10). The test result can be sent back to the user, operator, HMO, or physician for further confirmation, recording, and medical treatment decision/follow-up. The detection system can close the loop between the user of the test device and the HMO/physician for recording of the test result and the provision of medical treatment/follow-up.

Abstract: Aspects of the invention include includes detecting, using a first machine learning model, a first well pad at a first location based at least in part on a first set of data comprising spectral data describing a gas emission from the first location. Detecting an environmental event within a threshold distance of the well pad. Determining a probability of damage to the first well pad from the environmental event.

Abstract: Various examples are directed to systems and methods for tuning a database service in a cloud platform. A tuning service may access a neural network model trained to classify workload points to one of classes. The tuning service may execute the neural network model with a first source workload point as input to return a first class as output, where the first source workload describing a source database. The tuning service may select a target workload for the first source workload point from a plurality of reference workloads. Selecting the target workload may be based at least in part on the first class returned by the neural network model. The tuning service may generate a recommended knob configuration for the source database using the target workload.

Abstract: An image processing circuit capable of detecting an edge component includes: a selecting circuit acquiring the brightness values of pixels of an image according to the position of a target pixel and a processing region, wherein the pixels include N horizontal lines and M vertical lines; a brightness-variation calculating circuit generating N horizontal-line-brightness-variation values according to brightness variation of the N horizontal lines, and generating M vertical-line-brightness-variation values according to brightness variation of the M vertical lines; a brightness-variation determining circuit choosing a horizontal-line-brightness-variation representative value among the N horizontal-line-brightness-variation values, choosing a vertical-line-brightness-variation representative value among the M vertical-line-brightness-variation values, and choosing a brightness-variation representative value between the two representative values; an energy-variation calculating circuit generating an energy-variation

Abstract: Systems and methods are described, and an example method includes a training an artificial intelligence (AI) classifier of scanned items, including obtaining a training set of sample raw scans. The set includes a population of sample in-class raw scans, which include blocks of sensor data from scans of regions having in-class objects, and the set includes a population of sample not-in-class raw scans, which include blocks of sensor data from scan of regions without in-class objects. The example includes applying the AI classifier to sample raw scans in the training set, measuring errors in the results, and updating classifier parameters based on the errors, until detecting a training completion state.

Abstract: An apparatus includes a generation unit configured to generate shape information of an object in a captured image, a component acquisition unit configured to acquire an auxiliary light component representing intensity of an auxiliary light at each pixel of the captured image based on a light amount characteristic representing a light amount of the auxiliary light received by the object when the auxiliary light is emitted and the shape information of the object, a first correction unit configured to generate a first corrected image in which color of the captured image is corrected according to environmental light, a second correction unit configured to generate a second corrected image in which color of the captured image is corrected according to the auxiliary light, and a combining unit configured to combine the first corrected image and the second corrected image at a combination ratio calculated based on the auxiliary light component.

Abstract: A method for reducing abasing artefacts in an output image may include obtaining a plurality of input images captured by a plurality of cameras, each camera having a different field of view of an environment surrounding a vehicle, wherein the plurality of input images are mapped to the output image to represent the environment, from a predefined virtual point of view. The method may further include for each pixel position in the output image, obtaining a first pixel density value corresponding to a first output pixel position in the output image; and upon determining that the first pixel density value is higher than a threshold, calculating a first output brightness value corresponding to the first output pixel position based at least on a plurality of brightness values corresponding to a plurality of neighboring pixels of a corresponding position in a first input image of the plurality of input images.

Abstract: The invention belongs to the technical field of quantitative statistical distribution analysis for micro-structures of metal materials, and relates to a method for automatic quantitative statistical distribution characterization of dendrite structures in a full view field of metal materials. According to the method based on deep learning in the present invention, dendrite structure feature maps are marked and trained to obtain a corresponding object detection model, so as to carry out automatic identification and marking of dendrite structure centers in a full view field; and in combination with an image processing method, feature parameters in the full view field such as morphology, position, number and spacing of all dendrite structures within a large range are obtained quickly, thereby achieving quantitative statistical distribution characterization of dendrite structures in the metal material.

Abstract: A detection and tracking system and method using a camera unit on a robot, or alternatively a camera mounted inside the pool overlooking the bottom of the pool, for safety monitoring for use in and around water-related environments. The robot is able to propel itself and move throughout the body of water, both on the surface and underwater, and the camera unit functions both on the surface and underwater. The robot optimizes the cleaning cycle of the body of water utilizing deep learning techniques. The robot has localization sensors and software that allow the robot to be aware of the robot's position in the pool. The camera is able to send its video feed live over the internet, the processing is performed in the cloud, and the robot sends and receives data from the cloud. The processing utilizes deep learning algorithms, including artificial neural networks, that perform video analytics.

Abstract: Disclosed is a method for quantifying algal for management of water quality, performed by a computing device. The method may include: receiving a remote sensing image of an object of interest; and predicting a water quality variable based on the remote sensing image using a pre-trained algal estimation model.

Abstract: A device includes image generation circuitry and convolutional-neural-network circuitry. The image generation circuitry, in operation, generates a digital image representation of a wafer defect map (WDM). The convolutional-neural-network circuitry, in operation, generates a defect classification associated with the WDM based on the digital image representation of the WDM and a data-driven model generated using an artificial wafer defect digital image (AWDI) data set and associating AWDIs with classes of a defined set of classes of wafer defects. A wafer manufacturing process may be controlled based on the classifications of WDMs.

Abstract: Techniques for applying one or more machine learning models to a sub-region less than all of an image scene are described. An example is receiving first sub-region from an image; analyzing the received first sub-region of the image using the indicated least one machine learning model to perform the analyzing of the first sub-region of the scene; and outputting a result of the analyzing.

Abstract: A method for designing a material for an aircraft component includes training a neural network to correlate microstructural features of an alloy with material properties of the alloy by at least providing a set of images of the alloy to the neural network. Each of the images in the set of images has varied constituent compositions. The method further includes providing the neural network with a set of determined material properties corresponding to each image, associating the microstructural features of each image with the set of empirically determined data corresponding to the image, and determining non-linear relationships between the microstructural features and corresponding empirically determined material properties via a machine learning algorithm, receiving a set of desired material properties of the alloy for aircraft component, and determining a set of microstructural features capable of achieving the desired material properties of the alloy based on the determined non-linear relationships.

Abstract: Systems and methods are provided for generating textual embeddings by tokenizing text data and generating vectors to be provided to a transformer system, where the textual embeddings are vector representations of semantic meanings of text that is part of the text data. The vectors may be averaged for every token of the generated textual embeddings and concatenating average output activations of two layers of the transformer system. Image embeddings may be generated with a convolutional neural network (CNN) from image data, wherein the image embeddings are vector representations of the images that are part of the image data. The textual embeddings and image embeddings may be combined to form combined embeddings to be provided to the transformer system.

Abstract: A method of detecting the presence of a prostate cancer in a human subject comprising the steps of (a) obtaining a histologically normal prostate tissue sample from the patient and (b) quantifying the epithelial thickness or gland lumen roundness of the tissue, wherein an increase in epithelial thickness or a decrease in gland lumen roundness indicates the presence of prostate cancer or a prostate cancer field defect.

Abstract: An image processing apparatus includes an input unit that inputs an image, and a processor configured to read out a program stored in a memory, and executes the program. The processor is configured to detect an intended subject from the input image by a first detection method, set an intended subject region for the detected intended subject, detect the intended subject from the input image by a second detection method different from the first detection method, and update the set intended subject region by using a detection result of the second detection method.