Abstract: Mechanisms are provided to implement a hybrid deep learning network. The hybrid deep learning network receives, from a imaging system, first input data specifying a non-annotated image. The hybrid deep learning network pre-processes the non-annotated image to generate second input data specifying a hint image and corresponding annotation data specifying salient regions of the hint image. The hybrid deep learning network processes the first input data and second input data to perform training of the hybrid deep learning network by targeting feature detection in the non-annotated image in the salient regions identified in the hint image. The trained hybrid deep learning network is used to process third input data specifying a new non-annotated image to thereby identify an object or structure in the new non-annotated image.

Abstract: Systems, methods, and non-transitory computer-readable media can determine a landscape video to be presented in a vertical orientation through a display screen of a computing device. Frames of the landscape video can be cropped to allow for full screen presentation of the landscape video in the vertical orientation. The cropped frames of the landscape video can be provided for presentation through the display screen of the computing device in the vertical orientation.

Abstract: The present disclosure relates to a target, a method, and a system for calibrating a camera. One example embodiment includes a target. The target includes a first pattern of fiducial markers. The target also includes a second pattern of fiducial markers. The first pattern of fiducial markers is a scaled version of the second pattern of fiducial markers, such that a calibration image captured of the target simulates multiple images of a single pattern captured at multiple calibration perspectives.

Abstract: Implementations described herein discloses a multi-modality video recognition system. Specifically, the multi-modality video recognition system is configured to train a plurality of classifier networks, each of the classifier network trained with a different one of the plurality of video streams, wherein each of the plurality of different classifier networks includes multiple intermediate layers, determine correlation matrices of related intermediate layers of each of the plurality of the different classifier networks, and align the correlation matrices of the related intermediate layers of each of the plurality of the different classifier networks.

Abstract: A method for controlling a movable object includes obtaining current location information of an obstacle while the movable object tracks a target, and determining whether a location of the obstacle corresponds to a reactive region relative to the movable object based on the current location information of the obstacle. In response to determining that the location of the obstacle corresponds to the reactive region, one or more movement characteristics of the movable object is adjusted in a reactive manner to prevent the movable object from colliding with the obstacle. In response to determining that the location of the obstacle does not correspond to the reactive region, the one or more movement characteristics of the movable object is adjusted in a proactive manner to maintain a distance between the movable object and the obstacle to be larger than a predefined distance.

Abstract: A system and method for detecting a presence or absence of objects in a trailer are described. A 3D depth-camera is oriented to capture a 3D image. The 3D image includes a plurality of 3D point data defining a portion of a wall, floor, and top of a trailer. The plurality of 3D point data is then analyzed to determine a first, second, and third sub-plurality of points, associated with the portion of the wall, floor, and top, respectively. The first, second, and third sub-pluralities are then removed from the plurality of points to obtain a modified plurality of points, representing a modified 3D image. The modified 3D image is then segmented into a plurality of bins, and the plurality of bins are analyzed to determine one or more points-bin values. A communication is then provided based on whether any of the points-bin values exceeds a threshold value.

Abstract: To augment an image caption, a caption graph containing entity nodes corresponding to entities contained in the image and relationship edges between entity nodes corresponding to relationships between entities as illustrated in the image is generated. In addition, a contextual graph containing one or more of entity nodes corresponding to entities contained in the image and described in text associated with the image, textual entity nodes corresponding to textual entities described in text associated with the image and textual relationship edges between entity node pairs, textual entity node pairs and entity node and textual entity node pairs is generated. The textual relationship edges correspond to relationships described in the text associated with the image between entity pairs, textual entity pairs or entity and textual entity pairs. From the contextual graph, an augmented caption graph containing entity nodes, relationship edges, textual entities and textual relationship edges is generated.

Abstract: Certain examples described herein enable semantically-labelled representations of a three-dimensional (3D) space to be generated from video data. In described examples, a 3D representation is a surface element or ‘surfel’ representation, where the geometry of the space is modelled using a plurality of surfaces that are defined within a 3D co-ordinate system. Object-label probability values for spatial elements of frames of video data may be determined using a two-dimensional image classifier. Surface elements that correspond to the spatial elements are identified based on a projection of the surface element representation using an estimated pose for a frame. Object-label probability values for the surface elements are then updated based on the object-label probability values for corresponding spatial elements. This results in a semantically-labelled 3D surface element representation of objects present in the video data.

Abstract: Defects of interest can be captured by a classifier. Images of a semiconductor wafer can be received at a deep learning classification module. These images can be sorted into soft decisions with the deep learning classification module. A class of the defect of interest for an image can be determined from the soft decisions. The deep learning classification module can be in electronic communication with an optical inspection system or other types of semiconductor inspection systems.

Abstract: A system for capturing, using a digital camera, a Scrum board note and processing the image of the note to extract textual content of the note for use by a collaboration and project management application program. The processing of the image includes removing image imperfections and formatting the image such that it is optimized for scanning, thereby reducing spelling or scanning errors.

Abstract: Aspects of the disclosure relate to training a labeling model to automatically generate labels for objects detected in a vehicle's environment. In this regard, one or more computing devices may receive sensor data corresponding to a series of frames perceived by the vehicle, each frame being captured at a different time point during a trip of the vehicle. The computing devices may also receive bounding boxes generated by a first labeling model for objects detected in the series of frames. The computing devices may receive user inputs including an adjustment to at least one of the bounding boxes, the adjustment corrects a displacement of the at least one of the bounding boxes caused by a sensing inaccuracy. The computing devices may train a second labeling model using the sensor data, the bounding boxes, and the adjustment to increase accuracy of the second labeling model when automatically generating bounding boxes.

Abstract: A method of constructing a three-dimensional model of an internal surface of a tubular structure comprises obtaining image data from an area of the internal surface of the structure, obtaining measured height profile data from the internal surface in a plurality of sub-regions of the area, for example using a multi-finger caliper tool, determining image properties from the image data, correlating the measured height profile data with the image properties in the sub-regions, and constructing expected height profile data for at least part of the area outside the sub-regions using the correlated measured height profile data and image properties.

Abstract: Some embodiments provide methods for providing images of a person generated by two or more A/V recording and communication devices to one or more users, via a user's client device. For example, first image data may be received from a first A/V recording and communication device at a first location and second image data may be received from a second A/V recording and communication device at a second location. The first image data and the second image data may be analyzed to determine a person depicted in the first image data and a person depicted in the second image data is the same person. In response, a user alert may be generated including data representative of a first facial image of the person and a second facial image of the person. The user alert may then be transmitted to a user's client device.

Abstract: In some examples, luminance-biased sharpening for thermal media printing may include converting an input image to a grayscale luminance representation. For each pixel of a plurality of specified pixels of the converted input image, a sharpening lightness value may be determined. Further, for each pixel of the plurality of specified pixels of the converted input image, a ratio of the sharpening lightness value to a corresponding original lightness value may be determined. Based on the determined ratio, a sharpened output image corresponding to the input image may be generated.

Abstract: A nighttime vehicle detecting method is for capturing an image by a camera and driving a computing unit to compute the image and then detect a highlight point of the image. The nighttime vehicle detecting method is for driving the computing unit to perform a communicating region labeling algorithm to label a plurality of highlight pixels connected to each other as a communicating region value, and then performing an area filtering algorithm to analyze an area of the highlight pixels connected to each other and judge whether the highlight pixels connected to each other are a vehicle lamp or not according to a size of the area. The nighttime vehicle detecting method is for driving the computing unit to perform an optical flow algorithm to obtain a speed of the vehicle lamp, and then filtering the vehicle lamp moved at the speed smaller than a predetermined speed.

Abstract: Techniques are provided for recognition of activity in a sequence of video image frames that include depth information. A methodology embodying the techniques includes segmenting each of the received image frames into a multiple windows and generating spatio-temporal image cells from groupings of windows from a selected sub-sequence of the frames. The method also includes calculating a four dimensional (4D) optical flow vector for each of the pixels of each of the image cells and calculating a three dimensional (3D) angular representation from each of the optical flow vectors. The method further includes generating a classification feature for each of the image cells based on a histogram of the 3D angular representations of the pixels in that image cell. The classification features are then provided to a recognition classifier configured to recognize the type of activity depicted in the video sequence, based on the generated classification features.

Abstract: Embodiments of the disclosure provide methods and systems for calibrating a plurality of sensors. The method may include capturing, by a plurality of sensors associated with a vehicle, a set of point cloud data indicative of at least one surrounding object as the vehicle travels along a trajectory. The method may also include filtering, by a processor, the set of point cloud data based on coplanarity associated with the set of point cloud data. The method may further include adjusting, by the processor, at least one calibration parameter of the plurality of sensors based on a model using the filtered set of point cloud data. The model may include a weight corresponding to the coplanarity associated with the set of point cloud data.

Abstract: A method includes generating, based on first surface data that includes three-dimensional (3D) point positions corresponding to a first portion of a surface of an object, first hole data that indicates first positions of holes in the first portion of the surface. The method includes generating, based on second surface data corresponding to a second portion of the surface of the object, second hole data that indicates second positions of the holes in the second portion of the surface. The method also includes matching the first positions to the second positions to perform an alignment with respect to the first surface data and the second surface data.

Abstract: A method (200) for providing feedback regarding facial expressions in an original image is provided. The method (200) comprising the steps of receiving a 2-dimensional image of one or more users from an image capturing device (102), converting the 2-dimensional image to 3-dimensional image, identifying respective one or more faces of the one or more users from the 3-dimensional image, determining values of a set of plurality of parameters associated with a plurality of sections of each face of the one or more faces, comparing the determined values of the set of plurality of parameters with pre-determined optimum values of set of plurality of parameter pre-stored in a data repository (108) and providing feedback to the one or more users on the basis of the comparison. Further, a system (700) for providing feedback regarding facial expressions in an original image is provided.

Abstract: Embodiments of the present application provide a method and an apparatus, and a machine readable media for image processing. The method includes obtaining a panoramic video image, where the panoramic video image is determined based on a perspective mapping, and the panoramic video image includes a primary perspective region and at least one secondary perspective region; dividing the secondary perspective region into at least two sub-regions based on distribution information of high-frequency components in the secondary perspective region; determining respective filter templates of the sub-regions, and filtering the sub-regions using the respective filter templates; and determining a filtered panoramic video image.