Abstract: An example method includes acquiring a three-dimensional coordinate system associated with a three-dimensional camera and a first reference point associated with the three-dimensional camera, acquiring a two-dimensional coordinate system associated with a two-dimensional camera and a second reference point associated with the two-dimensional camera, aligning the three-dimensional coordinate system with the two-dimensional coordinate system, based on a fixed and known positional relationship between the first reference point and the second reference point, to obtain a fixed positional relationship between the three-dimensional coordinate system and the two-dimensional coordinate system, and transferring the three-dimensional coordinate system to a datum point of the two-dimensional coordinate system, using the fixed positional relationship between the three-dimensional coordinate system and the two-dimensional coordinate system.

Abstract: The present invention discloses a system and method for image processing and recognizing a scene of an image. The system utilizes a Multi-mode scalable network system and regrouping pipeline. The system is AI based system which uses neuro network. The system includes a pre-processing, processing and a post-processing unit. The system uses optical information recorded from the camera of a mobile device to extract and analyze the content in an image such as a photo or video clip. Based on the retrieved information, a label is given to best describe the scene of the image.

Abstract: Image features are extracted from multiple variably-illuminated images, with the images acquired from a multi-camera array microscope. A sequence of images per camera is captured, with the illumination pattern varied between each image capture. After image capture, a post-processing algorithm finds keypoints of interest within the images captured by each camera. These features can be used to assist with stitching together images from the multi-camera array, in addition to image compression, object tracking or other automated tasks.

Abstract: A mask structure optimization device includes a classification target image size acquisition unit that is configured to acquire a size of a classification target image which is an image including a classification target, a mask size setting unit that is configured to set a size of a mask applied to the classification target image, a brightness detection unit that is configured to detect a brightness of each pixel within the classification target image at a position on an opposite side of the mask from the classification target image, a sum total brightness calculation unit that is configured to calculate the sum total brightness of the each pixel within the classification target image detected by the brightness detection unit, an initial value setting unit that is configured to set an initial value for a mask pattern of the mask, and a movement unit that is configured to relatively move the mask with respect to the classification target image.

Abstract: Techniques for automatic image tagging and selection at a mobile device include generating a smart image tagging model by first training an initial model based on different angles of image capture of subject vehicles, and then re-training the trained model using weights discovered from the first training and images that have been labeled with additional tags indicative of different vehicle portions and/or vehicle parameters. Nodes that are training-specific are removed from the re-trained model, and the lightweight model is serialized to generate the smart image tagging model. The generated model may autonomously execute at an imaging device to predict respective tags associated with a stream of frames; select, capture and store respective suitable frames as representative images corresponding to the predicted tags; and provide the set of representative images and associated tags for use in determining vehicle damage, insurance claims, and the like.

Abstract: A system for detecting clouds and cloud shadows is described. In one approach, clouds and cloud shadows within a remote sensing image are detected through a three step process. In the first stage a high-precision low-recall classifier is used to identify cloud seed pixels within the image. In the second stage, a low-precision high-recall classifier is used to identify potential cloud pixels within the image. Additionally, in the second stage, the cloud seed pixels are grown into the potential cloud pixels to identify clusters of pixels which have a high likelihood of representing clouds. In the third stage, a geometric technique is used to determine pixels which likely represent shadows cast by the clouds identified in the second stage. The clouds identified in the second stage and the shadows identified in the third stage are then exported as a cloud mask and shadow mask of the remote sensing image.

Abstract: A processor-implemented method using a neural network (NN) includes: receiving input data; and determining information inferred from the input data based on state information about a state in which the NN is activated in response to the input data, wherein an embedding vector generated by encoding the input data using at least a portion of the NN comprises information used to reconstruct a first partial region of the input data with a first accuracy and to reconstruct a second partial region of the input data with a second accuracy, and wherein the first partial region is adaptively determined based on either one or both of the inferred information and the embedding vector.

Abstract: A method for all-optical reduction of inter-channel crosstalk for spectrally overlapped optical signals for maximizing utilization of an available spectrum includes receiving a plurality of spectrally overlapped optical signals modulated with data. The method further includes generating conjugate copies of each of the plurality of optical signals using non-linear optics. The method further includes selecting the conjugate copies and adjusting an amplitude, a phase, and a delay of the conjugate copies. The method further includes performing inter-channel interference (ICI) compensation on the spectrally overlapped optical signals in an optical domain by adding the adjusted conjugate copies to the spectrally overlapped optical signals.

Abstract: A first image is obtained prior to a second image. Upon dividing the second image into a plurality of zones, light characteristic values for each of the plurality of zones are determined based on at least one of a brightness value, a contrast value, and an intensity value for the respective zones. A first difference score for one of the zones is determined based on a difference between corresponding light characteristic values for the one zone in the first and second images. Upon identifying an object in the one zone of the first and second images, a second difference score for the one zone is determined based on determining a difference between respective confidence scores associated with the corresponding identified objects. Upon determining at least one of the first or second difference score is greater than a respective threshold, the vehicle is operated based on the first image.

Abstract: A processing device includes a first processor configured to detect a bounding box of a distant vehicle, in an input image generated by imaging the distant vehicle, and extract at least one feature of the distant vehicle. A second processor is configured to estimate a geometry of a road on which the distant vehicle is located, based on a position of at least one feature relative to at least a portion of the bounding box.

Abstract: Vehicle alert and control systems and methods taking into account a detected road friction at a following vehicle and a predicted road friction by the following vehicle. The detected road friction between the following vehicle tires and the road surface may be assessed using a variety of methodologies and is used to compute a critical safety distance between the following vehicle and the preceding vehicle and a critical safety speed of the following vehicle. The predicted road friction ahead of the following vehicle may also be assessed using a variety of methodologies (lidar, camera, and cloud-based examples are provided) and is used to compute a warning safety distance between the following vehicle and the preceding vehicle and a warning safety speed of the following vehicle. These functionalities may be applied to vehicle/stationary object warning and response scenarios as well.

Abstract: An apparatus including a memory and a circuit. The memory may comprise three buffers. The circuit may be configured to allocate the three buffers in the memory based on a size of a full resolution feature map, receive a plurality of regions of interest ranked based on a feature map pyramid, generate a plurality of levels of the feature map pyramid starting from the full resolution feature map and store the levels in the buffers. The circuit may store the levels that are used by at least one of the plurality of regions of interest or do have a dependent level, the levels that are generated may be stored in the buffers in a pattern that ensures the level is stored until no longer needed to create the dependent level and enables the level to be discarded when no longer needed to create the dependent level.

Abstract: When an object is detected by an electronic device in a predefined spatial region of a physical mat, the electronic device may perform one or more measurements of the object using the one or more sensors. Note that the physical mat may be on a surface that is separate from the electronic device. Then, the electronic device may identify the object based at least in part on the one or more measurements, where the identification involves a pretrained neural network or a pretrained machine-learning model that uses the one or more measurements as an input and that outputs information specifying the identified object. Moreover, the electronic device may provide classification information associated with the identified object. Next, the electronic device may perform analysis associated with the identified object. For example, the electronic device may increment a count of a number of a type of object that includes the identified object.

Abstract: In the present invention, a right leg drive RLD monitoring system is employed on a medical computing system/computer, such as an ECG, HEMO and/or EP monitoring, mapping and/or recording system, that includes a number of RLD circuits to be utilized for different procedures or monitoring states to be performed using the system. The RLD monitoring system operates to actively monitor and/or record the feedback voltage to the RLD isolated from the patient. Using the measured feedback voltage data, the RLD monitoring system can identify and determine if the RLD circuit in use is approaching saturation, has reached saturation and the duration the RLD circuit was in saturation. The RLD monitoring system can concurrently and/or subsequently select and/or provide selection information regarding an optimal RLD circuit to be utilized to most effectively perform the desired function of the RLD in the procedure being performed using the monitoring, mapping and/or recording system.

Abstract: Provided are a method and an apparatus for measuring an endolymphatic hydrops ratio of inner ear organs using an artificial neural network. The method of measuring an endolymphatic hydrops ratio includes obtaining a plurality of frame images obtained by capturing inner ear organs, obtaining a plurality of pieces of mask data corresponding to each of the plurality of frame images by inputting the plurality of frame images into a neural network, clustering the plurality of pieces of mask data according to the inner ear organs and obtaining representative images according to the inner ear organs according to certain conditions, and overlapping a target image synthesized by using the plurality of frame images and the representative images according to the inner ear organs so as to measure an endolymphatic hydrops ratio.

Abstract: This description relates to image feature extraction. In some examples, a system can include a keypoint detector and a feature list generator. The keypoint detector can be configured to upsample a keypoint score map to produce an upsampled keypoint score map. The keypoint score map can include feature scores indicative of a likelihood of at least one feature being present at keypoints in an image. The feature list generator can be configured to identify a subset of keypoints of the keypoints in the image using the feature scores of the up sampled keypoint score map, determine descriptors for the subset of keypoints based on a feature description map, and generate a keypoint descriptor map for the image based on the determined descriptors.

Abstract: Novel tools and techniques are provided for implementing visual impairment detection for free-space optical communication (“FSOC”) systems. In various embodiments, a computing system might receive, either from a camera or from a database, images (and/or videos) of an optical field of view (“FOV”) of the camera, the optical FOV comprising an optical beam(s) of a first FSOC system; might autonomously analyze the captured images (and/or videos) to determine whether an object(s) is moving within proximity to an optical beam(s) of the first FSOC system, to perform at least one of reactive learning or proactive learning regarding potential interruption of the optical beam(s) of the first FSOC system, and/or to determine one or more preventative measures to prevent interruption of the optical beam(s) of the first FSOC system; and might autonomously initiate one or more first tasks and/or the one or more preventative measures, based on the analysis.

Abstract: Method, systems and apparatuses may provide for technology that extracts one or more motion features from filtered position data associated with a projectile in a game and identifies a turning point in a trajectory of the projectile based on the one or more motion features. The technology may also automatically designate the turning point as a highlight moment if one or more of the turning point or the trajectory satisfies a proximity condition with respect to a target area in the game.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer readable media for accurately and efficiently learning parameters of a distilled neural network from parameters of a source neural network utilizing multiple augmentation strategies. For example, the disclosed systems can generate lightly augmented digital images and heavily augmented digital images. The disclosed systems can further learn parameters for a source neural network from the lightly augmented digital images. Moreover, the disclosed systems can learn parameters for a distilled neural network from the parameters learned for the source neural network. For example, the disclosed systems can compare classifications of heavily augmented digital images generated by the source neural network and the distilled neural network to transfer learned parameters from the source neural network to the distilled neural network via a knowledge distillation loss function.

Abstract: Methods, apparatus, systems, and articles of manufacture are disclosed that improve detection of malware based on ecosystem specific data. An example apparatus includes a feedback weight controller to apply, with a machine learning model, a weight to feedback associated with a sample, the feedback obtained from at least a customer ecosystem and including endpoint feedback, human feedback, infrastructure feedback, and global feedback; and a sample conviction controller to, in response to a score based on the weighted feedback satisfying a threshold for a classification, indicate to a user, with the machine learning model, that the classification for the sample is malicious.