Abstract: Iterative content-adaptive online training for end-to-end (E2E) neural image compression (NIC) using a neural network performed by at least one processor, is provided, including receiving an input image, to an E2E NIC framework, fine-tuning the E2E NIC framework, based on the input image, computing parameter updates using a first neural network of the fine-tuned E2E NIC framework, enhancing the fine-tuned E2E NIC framework based on a second neural network, the second neural network being a post-enhancement network, and generating an updated E2E NIC framework, based on the enhanced E2E NIC framework and the parameter updates.

Abstract: Embodiments of the present disclosure provide a method, an electronic device, and a computer program product for data processing. A method for data processing includes acquiring an image input and text input pair, the image input and text input pair comprising image input and text input, obtaining an image feature codebook corresponding to the image input and a text feature codebook corresponding to the text input, and training a text-image association model by using the image feature codebook and the text feature codebook, the text-image association model implementing the data processing based on the association between the image input and the text input. For example, a model may be adopted to learn an association between an image feature codebook corresponding to image input and a text feature codebook corresponding to text input, and data processing, such as data compression or data retrieval, is implemented using the learned association.

Abstract: In an embodiment a computer system includes at least one master module configured to process data having a format of N bits, a framebuffer configured to store pixel color component values of an image, the framebuffer having a resolution of N bits, each pixel being coded on P bits in the framebuffer and the pixels being stored one after another in the framebuffer and a memory management unit configured to control memory accesses of the at least one master module to the framebuffer, wherein the memory management unit is further configured to receive read memory access requests from the at least one master module, read at least one pixel in the framebuffer saved on P bits, and modify the format of the at least one read pixel by adding Q additional bits equal to a difference between N and P so as to format the at least one pixel on N bits before transmitting the at least one pixel to the at least one master module.

Abstract: A method for wafer image equalization includes obtaining a wafer image, converting the wafer image into bitmap data, generating a grayscale distribution of RGB pixels according to the bitmap data, generating a grayscale cumulative probability distribution of the RGB pixels according to the grayscale distribution, generating a mapping function according to the grayscale cumulative probability distribution of the RGB pixels, converting the grayscale distribution of the RGB pixels by the mapping function into an equalized grayscale distribution of the RGB pixels, and generating an equalized wafer image according to the equalized grayscale distribution of the RGB pixels.

Abstract: In one embodiment, a device receives a request for a machine learning model to make an inference about input data included in the request. The device retrieves metadata regarding training data used to train the machine learning model from a ledger associated with the machine learning model. The device assesses bias of the machine learning model by comparing the input data in the request to the metadata from the ledger. The device provides an indication of the bias of the machine learning model for display.

Abstract: Methods of training a machine learning model for image processing are described, as well as methods for image processing to determine a predicted rotation value for an image based on the trained machine learning model. A method of training includes utilising as a learning objective a reduction or minimisation of a classification loss. Training image pairs may be associated with classifications corresponding to rotation values for the purpose of determining the classification loss. An image may be divided into a plurality of images for determining an input to the machine learning model, for training or for image processing.

Abstract: A method of generating a plurality of spatially co-registered data elements, each spatially co-registered data element being associated with and generated from a pair of co-registered tissue sections obtained from adjacent positions of a core taken from a tissue sample and including an image data section and a genomic data section. The method includes, for each pair of co-registered tissue sections: (i) obtaining and storing as part of a data element a plurality of multi to hyperplexed images from the imaging data section of the co-registered tissue section, (ii) generating and storing as part of the data element image data from the plurality of multi to hyperplexed images, and (iii) generating and storing as part of the data element genomic data from the genomic data section of the associated co-registered tissue section.

Abstract: A computer-implemented method includes receiving an image of an article of interest to be evaluated relative to one or more features of interest, identifying a reference feature of a known size in the received image, identifying two or more extremities of the reference feature in the received image and a number of pixels between the two or more extremities of the reference feature, calculating a pixel size for the selected image based on the reference feature size and the number of pixels between the two or more extremities of the reference feature, annotating the received image to include one or more tolerance lines for the one or more features of interest, and determining whether the one or more features of interest in the image comply with the one or more tolerance lines. A computer program product and computer system corresponding to the method are also disclosed.

Abstract: Systems and methods used to perform touchless 3D scanning of a patient's face in the prone position are disclosed. In some embodiments, the systems include a rotatable 3D scanning device to capture 3D spatial data points of the patient's face and a reference frame. 3D digital mesh models are generated from the 3D spatial data points. A patient 3D digital mesh model is registered with a patient 3D model and a reference frame digital mesh model is registered with a reference frame 3D model. The 3D scanning device can include a handheld rotatable 3D scanner member or a mechanically rotatable 3D scanner member.

Abstract: Embodiments of the present application provide a medical image synthesis device and method. According to an embodiment, a method includes acquiring a first medical image and a second medical image and registering the first medical image with the second medical image. The method includes determining a first parameter value at each pixel location on the registered first medical image and a second parameter value at each pixel location on the second medical image. The method includes multiplying the first parameter value with the second parameter value at the same pixel location on the registered first medical image and the second medical image and generating synthetic image data based on the multiplication result.

Abstract: The present disclosure describes systems and methods for acquiring structural attributes of a building. One such method comprises acquiring, by one or more cameras of an unmanned aerial vehicle, infrared thermal and visible RGB aerial images of a building; identifying structural components of the building based on an infrared thermal aerial image, wherein a floor structure of the building is identified from a color change in the infrared thermal aerial image between the building and a foundation structure of the building; positioning a bounding box around the floor structure of the building present in the infrared thermal aerial image; and estimating a lowest floor elevation of the building by calculating a lowest floor elevation value based on an amount of image pixels representing a distance between a top of the bounding box and a top of the foundation structure of the building. Other methods and systems are also provided.

Abstract: A system and techniques for detecting a reflection of an object in a sequence of image frames. Objects of a given type are detected in the sequence of image frames, and a detection score is determined for each detected object. Distance ratios between unfiltered object position indicators identified for each detected object are determined. A displacement factor is determined for each detected object between a current location of the detected object and a previous location of the detected object. One of the two detected objects which has a lower detection score is a reflection of the other of the two detected objects is determined in response to finding a match between normalized sizes and normalized movements of the detected objects.

Abstract: An object recognition device being able to accurately specify a position and pose of a target object is provided. The object recognition device includes: a specification unit that specifies an object region including a three-dimensional point cloud indicating a position of a surface of a target object, based on a three-dimensional point cloud generated by converting a distance image including distance information to the target object, and specifies a reference shape similar to the target object, based on a three-dimensional point cloud included in the object region and feature information about a reference shape; and an estimation unit that estimates a reference position and a pose of the target object, based on a three-dimensional point cloud included in the object region and the specified reference shape.

Abstract: A method performed in a decoder includes receiving a coded video bitstream that includes (i) one or more two dimensional (2D) meshes corresponding to a three dimensional (3D) mesh and (ii) a 2D to 3D index array that maps each vertex in the one or more 2D meshes to a respective vertex in the 3D mesh. The method further includes reconstructing the 3D mesh using the 2D to 3D index array to map each vertex in the one or more 2D meshes to the respective vertex in the 3D mesh. The 2D to 3D index array encoded in a tuple format in which each tuple in the 2D to 3D index array includes a first parameter that specifies a starting index of a run, a second parameter that specifies a length of the run, and a third parameter that specifies a direction of the run.

Abstract: A method for detecting obstacles includes: acquiring a monitored image of a monitored region, performing object detection on the monitored image, performing pedestrian detection on the monitored image, in response to acquiring an object identification group comprising at least one object identification box by the object detection, and acquiring at least one pedestrian identification box by the pedestrian detection, acquiring a to-be-determined object identification group by removing an object identification box in the object identification group associated with the at least one pedestrian identification box, wherein the object identification box comprises an image of an object in the monitored image, and the pedestrian identification box comprises an image of a pedestrian in the monitored image, and in response to determining that an object identification box is present in the to-be-determined object identification group, determining that an obstacle is present in the monitored region.

Abstract: A computer-implemented segmentation method for segmenting a core and a penumbra in a four-dimensional computed tomography perfusion dataset of ischemic tissue in an image region of a patient, includes determining at least one parameter map for at least one perfusion parameter from the computed tomography perfusion dataset; and using the at least one parameter map and the computed tomograph perfusion dataset as input data to a trained function to determine output data, the output data including segmentation information of the penumbra and the core.

Abstract: A target tracking method, including: acquiring a video stream of a preset acquisition region; performing target detection on tth video frame of the video stream, and determining a first candidate box of the candidate target in the tth video frame; ranking the candidate targets, according to a confidence of the candidate target and a first intersection over union between the first candidate boxes of respective candidate targets, and determining a candidate target in at least one matched rank; performing at least one matching on the candidate target in the at least one matched rank, respectively, according to a set of trackers of the video stream in (t?1)th state and a preset matching strategy, and determining a target tracking result of the tth video frame; and determining a tracking trajectory of the target in the video stream according to the target tracking results of the T video frames.

Abstract: An image processing device includes a detection unit, a feature amount generating unit, a tracking unit, a feature amount change detecting unit, a tracking history extracting unit, and a determination unit, so that when occlusion in a selected cluster has occurred, the image processing device can determine whether or not objects are identical without using data of an image in which the occlusion has occurred. An image in which the occlusion has occurred can be excluded by using the data of the image determined to have no change in the feature amount, whereby a decrease in accuracy of determining whether or not the objects are identical can be suppressed.

Abstract: In one embodiment, a method includes accessing a video captured by cameras which is associated with a first framerate lower than a threshold framerate, for any two adjacent frames of the accessed video: generating a warped frame from the two adjacent frames based on an optical flow associated with the two adjacent frames, determining alignments for the two adjacent frames, respectively, fusing the determined alignments for the two adjacent frames, and generating a reconstructed frame based on the fused alignment, and reconstructing the accessed video based on the any two adjacent frames and their respective reconstructed frames, wherein the reconstructed video is associated with a second framerate higher than the threshold framerate.

Abstract: The present disclosure relates to methods and systems for vascular image processing. The method may include obtaining an initial vascular image, generating a vascular fragment image by performing a vascular fragmentation operation on the initial vascular image, and generating, based on the vascular fragment image, a vascular centerline image.