Abstract: A computer-implemented method of machine-learning is described that includes obtaining a test dataset of scenes. The test dataset belongs to a test domain. The method includes obtaining a domain-adaptive neural network. The domain-adaptive neural network is a machine-learned neural network taught using data obtained from a training domain. The domain-adaptive neural network is configured for inference of spatially reconfigurable objects in a scene of the test domain. The method further includes determining an intermediary domain. The intermediary domain is closer to the training domain than the test domain in terms of data distributions. The method further includes inferring, by applying the domain-adaptive neural network, a spatially reconfigurable object from a scene of the test domain transferred on the intermediary domain. Such a method constitutes an improved method of machine learning with a dataset of scenes comprising spatially reconfigurable objects.

Abstract: An electronic device is provided. The electronic device includes: a memory configured to include at least one instruction; and a processor configured to be connected to the memory to control the electronic device, and obtain an output image by upscaling an input image using an artificial intelligence model trained to upscale an image, wherein the processor is configured to control the electronic device to: obtain parameter information of the artificial intelligence model based on pre-processing related information performed on the input image, and upscale the input image using the artificial intelligence model corresponding to the obtained parameter information.

Abstract: Systems, devices, methods, and computer-readable media for determining planarity in a 3D data set are provided. A method can include receiving or retrieving three-dimensional (3D) data of a geographical region, dividing the 3D data into first contiguous regions of specified first geographical dimensions, determining, for each first contiguous region of the first contiguous regions, respective measures of variation, identifying, based on the respective measures of variation, a search radius, dividing the 3D data into respective second contiguous or overlapping regions with dimensions the size of the identified search radius, and determining, based on the identified search radius, a planarity of each of the respective second contiguous or overlapping regions.

Abstract: The image processing device comprises: a storage section storing a three-dimensional shape model in which feature amounts and three-dimensional positional information, for multiple feature points of a target object, are associated; an extraction process section configured to extract the feature amounts and two-dimensional positional information of the feature points from a two-dimensional image of the target object captured with a camera; and a recognition process section configured to identify three-dimensional positional information of the feature points of the two-dimensional image and recognize the position and orientation of the target object by matching the feature points of the two-dimensional image with the feature points of the three-dimensional model using the feature amounts.

Abstract: The method for item recognition can include: optionally calibrating a sampling system, determining visual data using the sampling system, determining a point cloud, determining region masks based on the point cloud, generating a surface reconstruction for each item, generating image segments for each item based on the surface reconstruction, and determining a class identifier for each item using the respective image segments.

Abstract: A method for optimizing a deep learning operator, includes: calling a method of reading an image object to read target data from an L1 cache of an image processor to the processor in response to detecting the target data in the L1 cache, performing a secondary quantization operation on the target data in the processor to obtain an operation result and writing the operation result into a main memory of the image processor. The target data is fixed-point data obtained after performing a quantization operation on data to be quantized in advance and the data to be quantized is one of the following: float-point data of an initial network layer of the neural network model and fixed-point data outputted from a network layer previous to the current network layer.

Abstract: To reduce the feel of incongruity in a model, provided is a detection device comprising: a texture detector that detects texture information of a target object from a first position; a position detector that detects depth information to each point in the target object from a second position different from the first position; a region detector that detects a data deficient region in which the depth information has been acquired but the texture information has not been acquired, on the basis of a detection result of the texture detector and a detection result of the position detector; and an adder that adds specific texture information to the data deficient region.

Abstract: Provided is a method of compressing an occupancy map of a three-dimensional (3D) point cloud, and more specifically, a method of compressing an occupancy map of a point cloud in which an occupancy map image of a point cloud existing in a 3D space is compressed based on a compression quality or a patch-by-patch inspection method of the occupancy map image so that compression distortion is minimized when reconstructing the compressed occupancy map image so as to remarkably improve the quality of a reconstructed occupancy map image.

Abstract: The present disclosure provides an image denoising method, an image denoising device, an image denoising apparatus and a storage medium. The method includes: performing edge detection on a color image to obtain a preprocessed image; acquiring a depth image having the same scene as the color image; and performing a first noise reduction process on the preprocessed image according to the depth image to obtain a first image.

Abstract: A method, computer program, or computer system is provided for compressing a neural network model. One or more blocks are identified from among a superblock corresponding to a multi-dimensional tensor associated with a neural network. A set of weight coefficients associated with the superblock is unified. A model of the neural network is compressed based on the unified set of weight coefficients.

Abstract: A signal processing apparatus comprises a decoding unit configured to generate a decoded image by decoding lossy compressed image data, and a restoration processing unit configured to perform image restoration processing on the decoded image. The restoration processing unit determines whether or not to perform the restoration processing for each of blocks in the decoded image in accordance with specific image information, and for a block on which it is determined that the restoration processing is to be performed, performs the restoration processing on the basis of an inference made using a coefficient learned in advance.

Abstract: In an image processing system in which development processing is applied to a RAW image in a server apparatus and results thereof are provided to a client terminal, the capacity of image data that is transmitted to the client terminal is reduced. The client terminal applies low-load development processing to an input RAW image and stores and displays the results image thereof. On the other hand, the server generates a difference image between both results images by applying the low-load and high-load development processing and provides the difference image to the client terminal. Then, based on the difference image, the client terminal reproduces a results image of the high-load development processing.

Abstract: In one aspect, an example method for generating a candidate image for use as backdrop imagery for a graphical user interface is disclosed. The method includes receiving a raw image and determining an edge image from the raw image using edge detection. The method also includes identifying a candidate region of interest (ROI) in the raw image based on the candidate ROI enclosing a portion of the edge image having edge densities exceeding a threshold edge density. The method also includes manipulating the raw image relative to a backdrop imagery canvas for a graphical user interface based on a location of the candidate ROI within the raw image. The method also includes generating, based on the manipulating, a set of candidate backdrop images in which at least a portion of the candidate ROI occupies a preselected area of the backdrop imagery canvas, and storing the set of candidate backdrop images.

Abstract: As agents move about a materials handling facility, tracklets representative of the position of each agent are maintained along with a confidence score indicating a confidence that the position of the agent is known. If the confidence score falls below a threshold level, image data of the agent associated with the low confidence score is obtained and processed to generate one or more embedding vectors representative of the agent at a current position. Those embedding vectors are then compared with embedding vectors of other candidate agents to determine a set of embedding vectors having a highest similarity. The candidate agent represented by the set of embedding vectors having the highest similarity score is determined to be the agent and the position of that candidate agent is updated to the current position, thereby re-identifying the agent.

Abstract: The present disclosure relates to a class-agnostic object segmentation system that automatically detects, segments, and selects objects within digital images irrespective of object semantic classifications. For example, the object segmentation system utilizes a class-agnostic object segmentation neural network to segment each pixel in a digital image into an object mask. Further, in response to detecting a selection request of a target object, the object segmentation system utilizes a corresponding object mask to automatically select the target object within the digital image. In some implementations, the object segmentation system utilizes a class-agnostic object segmentation neural network to detect and automatically select a partial object in the digital image in response to a target object selection request.

Abstract: A method and apparatus for calibrating an image capture device are provided. The method includes capturing one or more of a single or Multiview image set by the image capture device, detecting one or more calibration features in each set by a processor, initializing each of the one or more calibration parameters a corresponding default value, extracting one or more relevant calibration parameters, computing an individual cost term for each of the identified relevant calibration parameters, and scaling each of the relevant cost terms. The method continues with combining all the cost terms once each of the calculated relevant cost terms have been scaled, determining if the combination of the cost terms has been minimized, adjusting the calibration parameters if it is determined that that the combination of the cost terms has not been minimized, and returning to the step of extracting one or more of the relevant calibration parameters.

Abstract: A G-PCC coder is configured to receive the point cloud data, determine a final quantization parameter (QP) value for the point cloud data as a function of a node QP offset multiplied by a geometry QP multiplier, and code the point cloud data using the final QP value to create an coded point cloud.

Abstract: Methods and devices for encoding a point cloud. A bit sequence signaling an occupancy pattern for sub-volumes of a volume is coded using binary entropy coding. Contexts may be based on neighbour configuration and a partial sequence of previously-coded bits of the bit sequence. A determination is made as to whether to apply a context reduction operation and, if so, the operation reduces the number of available contexts. Example context reduction operations include reducing neighbour configurations based on shielding by sub-volumes associated with previously-coded bits, special handling for empty neighbour configurations, and statistics-based context consolidation.

Abstract: A system for generating a high resolution (HR) computed tomography (CT) image from a low resolution (LR) CT image is described. The system includes a first generative adversarial network (GAN) and a second GAN. The first GAN includes a first generative neural network (G) configured to receive a training LR image dataset and to generate a corresponding estimated HR image dataset, and a first discriminative neural network (DY) configured to compare a training HR image dataset and the estimated HR image dataset. The second GAN includes a second generative neural network (F) configured to receive the training HR image dataset and to generate a corresponding estimated LR image dataset, and a second discriminative neural network (DX) configured to compare the training LR image dataset and the estimated LR image dataset.

Abstract: An image processing method includes extracting a first region in a first image by inputting the first image to a pretrained neural network, upscaling a resolution of the first region by performing neural network-based super resolution processing on the first region, and upscaling a resolution of a second region in the first image from which the first region is excluded by performing interpolation on the second region.