Abstract: In a method and apparatus for generating a depth map, the method includes obtaining a LiDAR point cloud generated by a LiDAR, obtaining a 3D bounding box for at least a portion of the LiDAR point cloud, generating a first vehicle mask corresponding to the 3D bounding box from an image point cloud obtained by projecting coordinates of points of the LiDAR point cloud into an image coordinate system, and generating the depth map by removing distant points that are not included in the 3D bounding box among points included in the first vehicle mask.

Abstract: A handheld object recognition method includes obtaining location information of each of one or more detected objects in a to-be-recognized image; obtaining a first label of each detected object indicating a type of the detected object, wherein the type of the detected object is used to represent a handheld relationship of the detected object; obtaining a handheld object from the one or more detected objects based on the first label; and obtaining a recognition result of the handheld object in the to-be-recognized image based on the location information of the handheld object.

Abstract: Devices, systems, and methods for extraction of routing card information of an event system are described herein. In some examples, one or more embodiments include a computing device comprising a memory and a processor to execute instructions stored in the memory to receive a file of a routing card associated with a facility including the event system, convert the file to an image, extract facility information, map and routing information, and event device information from the image of the routing card, and generate an output file having the extracted data as metadata in the output file.

Abstract: This disclosure provides a deep learning method, apparatus, device, and medium for task joint optimization based on vascular centerline guidance, including: in a process of extracting a global image feature by an encoder, extracting a local image feature after performing data augmentation on a local image with morphological specificity based on the vascular centerline guidance, and fusing the local image feature with the global image feature to form a fused image feature; and forming a bi-directional feature correction process between decoders of different decoding branches. The vascular centerline guidance is used to provide gated feature optimization and computational efficiency adjustment for an encoding process. Specific morphological data of the vascular centerline guidance is used to form a control mechanism for modulating efficiency of local feature extraction. A scarcity problem of hemodynamic annotations has been alleviated and generalization ability of a model has been enhanced.

Abstract: The disclosed technology is generally directed to cloud model orchestration of constrained-environment-device data. In one example of the technology, on a first distributed system, on-cloud models are stored. For each on-cloud model, a corresponding set of reference embeddings is stored. Images obtained from sensors onboard at least one constrained-environment device that is external to the first distributed system are received. Image tiles are stored such that the image tiles include, for each image, evenly-spaced portions of the image. Image embeddings generated via an embedding-generation model from the image tiles are stored. A request that is associated with a first model of the on-cloud models is received. Using the set of reference embeddings for the first model and the image embeddings, for each image tile, whether the first model should be executed on the image tile is determined. The first model is executed on the determined image tiles.

Abstract: A set of pre-annotated medical images is received, and the received set is processed by automatically: training an AI-based uncertainty model using the received set as training data; processing medical images of the received set by determining classified segments and/or uncertainty regions in the medical images using the trained AI-based uncertainty model; selecting at least a part of the processed medical images including classified segments and/or uncertainty regions based on the processing result; and presenting the selected part of processed medical images to a human expert. Furthermore, a modified received set including additional annotations created by the human expert is received.

Abstract: Systems and methods are provided for detecting and classifying driving behaviors, as well as techniques for improving such systems and methods, which may include identifying geographic locations where anomalous rates of non-compliant driving behaviors are observed, such as in the vicinity of certain traffic signs, traffic lights, and road markings.

Abstract: A method of image processing includes receiving a set of images from a sensor, dynamically determining respective cell resolutions of respective cells in a bird's-eye-view (BEV) grid based on image content in the set of images, wherein at least two of the cells have different cell resolutions, and generating BEV image content based on the respective cell resolutions of the respective cells.

Abstract: System apparatus, article of manufacture, method and/or computer program embodiments are provided for detecting and monitoring compliance matters. An example method may include obtaining a first set of image data corresponding to an environment from one or more image sensors; processing the first set of image data using a first machine learning algorithm to yield annotated image data that identifies at least one object within the environment; processing the annotated image data using a second machine learning algorithm to determine one or more attributes associated with the at least one object within the environment; and generating, based on the one or more attributes, an evidence package corresponding to the at least one object within the environment, the evidence package being configured to facilitate a determination of one or more action items associated with the at least one object.

Abstract: An eyeball tracking method and a virtual reality device are disclosed. The eyeball tracking method is applied to the virtual reality device, the virtual reality device comprises at least two cameras, and shooting areas of the cameras are partially overlapped. The eyeball tracking method comprises the steps of: acquiring images captured by at least two cameras at a same moment; superimposing overlapping areas in the images to obtain a target image including an eyeball area; and performing eyeball tracking according to the target image. The accuracy of eyeball tracking can be improved.

Abstract: A method is for verifying an implemented neural network and includes providing a plurality of validation images, feeding the plurality of validation images to an original neural network, feeding the plurality of validation images to the implemented neural network, and determining an activation coverage of the original neural network using all of the validation images of the provided plurality of validation images. The method further includes comparing the determined activation coverage with a target coverage, and comparing respective output images of the original neural network resulting from the plurality of validation images, with respective output images of the implemented neural network resulting from the plurality of validation images, in order to verify the implemented neural network.

Abstract: A system and method for authenticating an image is given. The method encompasses capturing a first image comprising at least one first human object and a second ID image; detecting at least one face of one first human object and processing the first human object based on face attribute checks, face quality checks, and face manipulation checks; performing a liveness check of the identification document to crop and align the region of interest; detecting a face of the at least one second human object; performing ID quality checks and ID tampering checks on the region of interest of the identification document; determining the image face feature vector and an ID card face feature vector; determining a match score between the image face feature vector and the ID card face feature vector and a status indicating one of authentication success, authentication failure, or a requirement for manual review of the image.

Abstract: A Method for processing RGBIR (Red Green Blue Infrared) multi-frame HDR (High Dynamic Range) with luma includes capturing a plurality of RGBIR raw images with different exposure times by at least one camera, generating a plurality of luma images using a color filter array luma (CFAL) filter according to the plurality of RGBIR raw images, performing IR color correction to generate a plurality of corrected RGBIR raw images according to the RGBIR raw images, performing HDR fusion to generate a HDR RGBIR raw image according to the plurality of luma images and the plurality of corrected RGBIR raw images, generating gradients according to the HDR RGBIR raw image and the plurality of luma images, and performing interpolation to generate a result image according to at least the gradients and the HDR RGBIR raw image.

Abstract: Provided are apparatuses, a non-transitory computer-readable medium or media, for predicting of neurodevelopmental disorder using a fundus image of a subject. In certain aspects, disclosed a method including the steps of: receiving the fundus image; processing the fundus image to classify one or more features contained in the fundus image using a machine learning model; and predicting, based on a classification, whether the fundus image is indicative of presence of neurodevelopmental disorder in the subject, wherein the machine learning model has been trained on stored fundus images obtained from at least two group of subjects who have previously been diagnosed as having the neurodevelopmental disorder, the machine learning model is trained by processes comprising; extracting a first embedding vector and a second embedding vector from a first fundus image and a second fundus image, respectively; and maximizing a similarity between the first embedding vector and the second embedding vector.

Abstract: Provided is an apparatus for recognizing an object that includes an object inference module configured to process an original image captured by a camera module and generate an image of a size to be input to a machine learning inference model, wherein the object inference module includes the machine learning inference model, and outputs a result of recognition and classification of an object inferred through the machine learning inference model, and the machine learning inference model processes an input image to infer an object included in the input image.

Abstract: System and techniques are described herein for processing images to detect expressions of a subject. In one illustrative example, a method of recognizing facial expressions in one or more images includes obtaining, by a computing device, a first image of a person; obtaining expression information based on the first image and an anchor image associated with the person; and determining an expression classification associated with the first image based on the expression information.

Abstract: A method for operating a technical system and a technical system. The method includes providing for at least one class at least one class attribute comprising a description for members of the class, providing features characterizing a digital image, determining a class of the at least one class that classifies the digital image depending on the features, and determining at least one first attribute depending on the at least one class attribute provided for the class that classifies the digital image. The at least one first attribute includes an explanation for classifying the digital image with the class that classifies the digital image. The method further includes operating the technical system depending on the class that classifies the digital image and/or depending on the at least one first attribute.

Abstract: An image processing method including: obtaining a target feature and a reference feature, the target feature is a feature obtained by performing feature extraction on a target object in a target image, and the reference feature is a feature obtained by performing feature extraction on a reference object in a reference image; obtaining a migration matrix, the migration matrix is used for controlling the reference feature to be fused with the target feature at a designated weight; fusing the target feature and the reference feature according to the migration matrix to obtain a fused feature; and generating a migration result image according to the fused feature and the target image. An image processing apparatus, an electronic device, a storage medium and a program product are also disclosed.

Abstract: Apparatuses, systems, and techniques to track object movements in a video. In at least one embodiment, a rate, at which one or more neural networks is to identify one or more objects within the video, may be adjusted based, at least in part on one or more user-selected portions of a video frame.

Abstract: A reconstruction method for spectral image includes: obtaining a measurement image of an imaging object, and reconstructing and obtaining a spectral image of the imaging object according to the measurement image and a pre-calibrated sensing matrix. The spectral image includes spectral information at different position points of the imaging object. Through the method, defects of the time-consuming spectral image reconstruction method and the low resolution of the obtained spectral image in the related art can be overcame and the spectral image of the target imaging object can be obtained by reconstructing quickly and the obtained spectral image has high spatial resolution and no mosaic.