Abstract: The method and system for labeling 2D data using a 3D model provides processes that include generating Neural Radiance Fields (NeRF) of a desired scene, converting the NeRF into a mesh, rendering images of the mesh in a simulator from a user-chosen new camera pose including position and orientation, labeling the newly rendered images where bounding boxes for the images are created, rendering images using the NeRF from the same user-chosen new camera pose, and overlaying the bounding boxes from the simulator onto the imagery from the NeRF. The method and system provide capability of augmenting a set of real images to create a more diverse training dataset that captures the scene and its objects from various new camera positions, while maintaining high qualities of images.

Abstract: A method of environment mapping includes the steps of receiving a plurality of images from respective different viewpoints; using a simultaneous localisation and mapping ‘SLAM’ process, detect feature points in the plurality of images; classifying one or more objects within the plurality of images as being objects of interest; determining the position of the or each object of interest using feature points corresponding to the objects of interest in the images; and providing an indication of the position of one or more objects of interest to a user of a virtual display; and where the step of determining the position of the or each object of interest using their feature points is repeated, more frequently than for other feature points in the images.

Abstract: A model generation method for generating a video extraction model for extracting a matching interval in a video that matches contents of an input sentence is provided. In the model generation method, a base matching interval and a sub matching interval in a training video are extracted by inputting a base sentence and a sub sentence to the video extraction model. Next, a loss for each of ground truth interval, the base matching interval, and the sub matching interval is calculated by processing a learning task of reconstructing the base sentence based on a feature value of the training video corresponding to each of them. Then, a machine learning is performed such that the first loss related to the ground truth interval is smaller than the second loss related to the base matching interval, and the second loss is smaller than the loss related to the sub matching interval.

Abstract: This application provide an image processing method, and relate to the field of computer vision technologies, to perform repair processing on a color image in a fog/haze/smoke extreme weather condition. The image processing method includes: obtaining a first depth map and a color image, where the first depth map and the color image are images respectively shot by a depth sensor and a camera for a same scene; determining an atmospheric light value; performing depth completion on the first depth map to obtain a second depth map, and determining a transmittance based on the second depth map; and performing repair processing on the color image based on the atmospheric light value and the transmittance. The second depth map is used to determine the transmittance, and therefore, a transmittance of higher precision is obtained, thereby improving an image processing effect.

Abstract: An image processing apparatus is to: detect a first edge point group comprising a set of edge points between a document background area and a document area in a read image of a document; calculate feature amounts from a plurality of edge points in the first edge point group and compose an edge point feature amount group; calculate a distribution of the feature amounts of the plurality of edge points from the edge point feature amount group and calculate a feature amount corresponding to a maximum distribution; select edge points corresponding to the maximum distribution from the first edge point group and select a second edge point group using a range of the feature amount corresponding to the maximum distribution; and calculate an edge straight line approximating the second edge point group.

Abstract: An area of Barrett's oesophagus in a subject's oesophagus is quantified from a video image signal representing a video image of the subject's oesophagus captured using a camera of an endoscope. Depth estimation on the frames to derive depth maps in respect of frames of the video image. Regions of the frames corresponding to an area of Barrett's oesophagus in the subject's oesophagus are segmented. A value of a geometrical measure of the area of Barrett's oesophagus is calculated using the depth map and segmented region in respect of at least one of the frames.

Abstract: A method for determining a recurrence of a disease in a patient includes generating a medical image of an organ of the patient and then extracting an invasive edge around an area of interest in the medical image. A plurality of radiomics features is obtained from the invasive edge and the recurrence of the disease is determined based on the plurality of radiomics features.

Abstract: The present disclosure relates to systems, methods, and computer-readable media for selectively identifying image frames from an input video to provide to an image processing model based on camera statistics. For example, systems disclosed herein include receiving an input video and associated camera statistics from a video capturing device. The systems disclosed herein further include identifying select image frames to provide to the image processing model based on the camera statistics and based on an application of the image processing model. The systems disclosed herein further include selectively identifying and providing camera statistics to the image processing model. By selectively providing data to the image processing model based on camera statistics, the systems disclosed herein can leverage capabilities of video capturing devices to significantly reduce the expense of processing resources when utilizing a variety of image processing models.

Abstract: A method includes: receiving data signals from a plurality of pixels of an array of pixels; generating a plurality of signal-to-noise ratios by determining signal-to-noise ratios for each respective pixel of the plurality of pixels on the basis of the data signals received from the respective pixel; and filtering the data signals received from each pixel of the plurality of pixels by using an adaptive filter configured on the basis of the plurality of the signal-to-noise ratios to generate filtered data signals.

Abstract: According to various embodiments, an electronic device may comprise: a memory; a display; a plurality of cameras disposed below the display; and at least one processor operatively connected to the memory, the display, and the plurality of cameras, wherein the at least one processor is configured to, while controlling a plurality of pixel areas of the display, which correspond to the plurality of cameras, acquire a plurality of images of a user through the plurality of cameras, obtain a final image based on at least one image among the plurality of images, and authenticate the user based on the final image.

Abstract: Provided are an electronic devices and methods for processing an image through a canonical space, the method including receiving training images of a scene obtained from a first sensor, generating first conversion images by converting the training images into images corresponding to a scene obtained through a second sensor, generating second conversion images by converting a color value of the training images and a color value of the first conversion images, and training a canonical image generative model configured to convert the training images, the first conversion images, and the second conversion images into canonical space images of a canonical color space and a canonical illumination space.

Abstract: The present subject matter refers a method of image processing through an artificial neural network implemented in an adapter card in a host-computing system.

Abstract: A robotic system is configured to automatically identify surgical instruments used during a bronchoscopy procedure. The robotic system can include a video capture device, a robotic manipulator, sensors configured to detect a configuration of the robotic manipulator, and control circuitry communicatively coupled to the robotic manipulator. The control circuitry is configured to perform, using a machine learning classifier, a first analysis of a bronchoscopy video of a patient site to track a medical instrument in the bronchoscopy video. The control circuitry can then identify a set of possible instrument identifications for the medical instrument in the bronchoscopy video based on the first analysis and an identified phase of the bronchoscopy procedure. The control circuitry can then track a motion of the medical instrument in the bronchoscopy video and select an identification from the set of possible instrument identification for the medical instrument based at least on the tracked motion.

Abstract: The present invention relates to a computer-implemented method capable of automatically detecting small bowel and colonic ulcers and erosions in Crohn's capsule endoscopy image data, by classifying pixels as lesion or non-lesion, using a convolutional image feature extraction step followed by a classification step and indexing such lesions in the set of one or more classes.

Abstract: Input data including a facsimile representation of a paper document is input at first branch of a zero-shot configuration (Zcon), which includes a first Vision-Language Model (VLM) adapted to identify a first set of fields in the input data using a visualized first set of bounding boxes (BB). A first agentic AI in the first branch localizes a field as a desired type of filed using a visualized identified BB. A second VLM in a second branch identifies a second set of fields in the input data, and outputs a visualized second set of BBs. The input data with the second set of BBs is passed to a MLLM executing in the second branch, which outputs a set of recognizing fields within the BBs of the second set of BBs. A second agentic AI in the second branch localizes at least one recognized field as a target field.

Abstract: Provided are a display apparatus and a method thereof. The display apparatus includes a display; a memory storing one or more instructions; and a processor configured to execute the one or more instructions stored in the memory to: receive an input image, obtain luminance information from the input image, obtain, based on the luminance information, perceptual luminance information in view of a visual response, predict scene intensity information indicating a degree of a side effect that occurs after a first image quality processing is performed on the input image, and perform the first image quality processing on the input image based on the scene intensity information and the perceptual luminance information.

Abstract: A method of initializing and training a neural network having a plurality of layers includes defining a first function configured to generate a filter based on a plurality of variance values associated with respective pairs of parameters of the plurality of layers, calculating the plurality of variance values based on depths of respective layers of the plurality of layers such that the variance values increase as the depths increase, calculating a covariance matrix using the first function, the covariance matrix having a block structure and each block of the covariance matrix corresponding to a covariance between a respective parameter and other parameters of the plurality of layers, providing, as input, the covariance matrix to the neural network to initialize the neural network for training, and generating, using the neural network, an output based on the covariance matrix.

Abstract: In a method for designing a binary phase filter to expand depth of field and for reconstructing images performed by an electronic device according to an embodiment of the present disclosure, the method includes: obtaining images for each depth for a test image based on a binary phase structure set in the binary phase filter; obtaining reconstructed images for each depth by reconstructing the images for each depth based on a model trained to reconstruct an input image; calculating a loss value of the reconstructed images for each depth by comparing the reconstructed images for each depth with the test image; and updating the binary phase structure until the loss value becomes less than or equal to a predefined value.

Abstract: The present invention provides an integrated hyperspectral water quality analysis method, which belongs to the field of hyperspectral water quality analysis. First, data preprocessing is conducted by water quality data collection and water quality image collection in early stage; second, three dimensionality reduction methods are adopted to conduct dimensionality reduction processing, and fused dimensionality reduction is conducted by parameter trade-off selection; third, machine learning algorithms are adopted to train and test hyperspectral water quality inversion models on spectral data after dimensionality reduction; finally, the hyperspectral water quality inversion models are selected and optimized. The present invention adopts an innovative fusion strategy in the aspect of data dimensionality reduction processing, which can achieve a better data dimensionality reduction effect, effectively remove noise and redundant information, and provide a more accurate and reliable data basis.

Abstract: A method for inverse tone mapping includes obtaining a histogram of a low dynamic range image, called LDR image and obtaining an ITMO function, allowing obtaining a pixel value of a High Dynamic Range image, called HDR image, from a pixel value of the LDR image and a gain function depending on said pixel value of the LDR image. A search process is applied using the obtained histogram to identify areas of the LDR image producing bright areas in the HDR image when the ITMO function is applied on said LDR image. Information representative of the bright areas is used to determine when modifying the gain function to ensure the HDR image respects at least one predefined light energy constraint.