Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium. In one aspect, a method includes receiving a text prompt describing a scene; processing the text prompt using a text encoder neural network to generate a contextual embedding of the text prompt; and processing the contextual embedding using a sequence of generative neural networks to generate a final video depicting the scene.

Abstract: A computer-implemented method is provided. The computer-implemented method includes inputting a low-resolution image into a generator; and generating a high-resolution image using the generator based on the low-resolution image. Generating the high-resolution image includes processing the low-resolution image through a plurality of super-resolution generating units arranged in series in the generator. A respective output from a respective one of the plurality of super-resolution generating units has a respective increased image resolution as compared to a respective input to the respective one of the plurality of super-resolution generating units.

Abstract: A point cloud encoding method is provided to improve coding efficiency. The method includes: obtaining global matched patches in N frames of point clouds in a point cloud group; determining M union patch occupancy maps corresponding to M sets, where a union patch occupancy map corresponding to an mth set is a union set of occupancy maps of all global matched patches in the mth set; packing the M union patch occupancy maps to obtain a global occupancy map; packing each of the N frames of point clouds to obtain occupancy maps of the N frames of point clouds; and encoding the N frames of point clouds based on the occupancy maps of the N frames of point clouds.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network. One of the methods includes receiving a training image and a ground truth super-resolution image; processing a first training network input comprising the training image using the neural network to generate a first training super-resolution image; processing a first critic input generated from (i) the training image and (ii) the ground truth super-resolution image using a critic neural network to map the first critic input to a latent representation; processing a second critic input generated from (i) the training image and (ii) the first training super-resolution image using the critic neural network to map the second critic input to a latent representation; determining a gradient of a generator loss function that measures a distance between the latent representations of the critic inputs; and determining an update to the parameters.

Abstract: Method and systems are provided for robust disparity estimation based on cost-volume attention. A method includes extracting first feature maps from left images captured by a first camera; extracting second feature maps from right images captured by a second camera; calculating a matching cost based on a comparison of the first and second feature maps to generate a cost volume; generating an attention-aware cost volume from the generated cost volume; and aggregating the attention-aware cost volume to generate an output disparity.

Abstract: An image processing apparatus applies an image to a first learning network model to optimize the edges of the image, applies the image to a second learning network model to optimize the texture of the image, and applies a first weight to the first image and a second weight to the second image based on information on the edge areas and the texture areas of the image to acquire an output image.

Abstract: A method for structure simulation for super-resolution fluorescence microscopy, the method including receiving a first image having a first resolution, which is indicative of a distribution of fluorophores; applying a Markov model to the fluorophores to indicate an emission state of the fluorophores; generating a plurality of second images, having the first resolution, based on the first image and the Markov model; adding DC background to the plurality of second images to generate a plurality of third images, having the first resolution; downsampling the plurality of third images to obtain a plurality of fourth images, which have a second resolution, lower than the first resolution; and generating a time-series, low-resolution images by adding noise to the plurality of fourth images. The time-series, low-resolution images have the second resolution.

Abstract: A method includes processing, using at least one processor of an electronic device, multiple reference images of a scene using a first convolutional neural network (CNN) to generate a confidence map and a disparity map. The method also includes generating, using the at least one processor, an initial Bokeh image based on the disparity map and the reference images using a depth-of-field (DoF) renderer. The method further includes refining, using the at least one processor, the initial Bokeh image using a second CNN to generate a refined Bokeh image, where the second CNN uses the confidence map, the disparity map, and the reference images to generate the refined Bokeh image.

Abstract: The current disclosure provides for mapping ultrasound images to resolution mapped ultrasound images using generative neural networks, while maintaining clinical quality of the resolution mapped ultrasound image, thereby enabling a clinician to evaluate ultrasound images in a preferred resolution without loss of clinically relevant content.

Abstract: The present disclosure includes methods and systems for identifying and manipulating a segment of a three-dimensional digital model based on soft classification of the three-dimensional digital model. In particular, one or more embodiments of the disclosed systems and methods identify a soft classification of a digital model and utilize the soft classification to tune segmentation algorithms. For example, the disclosed systems and methods can utilize a soft classification to select a segmentation algorithm from a plurality of segmentation algorithms, to combine segmentation parameters from a plurality of segmentation algorithms, and/or to identify input parameters for a segmentation algorithm. The disclosed systems and methods can utilize the tuned segmentation algorithms to accurately and efficiently identify a segment of a three-dimensional digital model.

Abstract: Provided is a method for training a neural network and a device thereof. The method may train a neural network with three-dimensional (3D) training image data including a plurality of two-dimensional (2D) training image data. The method may include training, at a processor, a first convolutional neural network (CNN) with the plurality of 2D training image data, wherein the first convolutional neural network comprises 2D convolutional layers. The method may further include training, at the processor, a second convolutional neural network with the 3D training image data, wherein the second convolutional neural network comprises the 2D convolutional layers and 3D convolutional layers configured to receive an output of the 2D convolutional layers as an input.

Abstract: An image reconstruction method, device and apparatus and non-transitory computer-readable storage medium are disclosed. The method may include: determining norms of convolution kernels of each convolutional layer of a deep neural network model; determining the convolution kernels with norms greater than or equal to a preset threshold in each convolutional layer to obtain a target convolution kernel set of each convolutional layer; processing an input image of each convolutional layer by using the convolution kernels in the target convolution kernel set of each convolutional layer respectively, to obtain a first image processing result; obtaining a second image processing result by performing interpolation on an initial image; and determining a fusion result according to the first image processing result and the second image processing result and reconstructing the initial image according to the fusion result.

Abstract: An electronic apparatus and a method for controlling the electronic apparatus are disclosed. The method includes: obtaining a neural network model trained to detect an object corresponding to at least one class; obtaining a user command for detecting a first object corresponding to a first class; and based on the first object not corresponding to the at least one class, obtaining a new neural network model based on the neural network model and information of the first object.

Abstract: Embodiments of this disclosure include an image enhancement method and apparatus. The image enhancement may include obtaining an original image and performing synthesis processing on features of the original image to obtain a first illumination map corresponding to the original image. A resolution of the first illumination map may be lower than a resolution of the original image. The image enhancement may further include obtaining, based on the first illumination map, a mapping relationship between an image to an illumination map and performing mapping processing on the original image based on the mapping relationship to obtain a second illumination map. A resolution of the second illumination map may be equal to the resolution of the original image. The image enhancement may further include performing image enhancement processing on the original image according to the second illumination map to obtain a target image.

Abstract: Various embodiments use a neural network to analyze images for aspects that characterize the images, to present locations of those aspects on the images, and, additionally, to permit a user to interact with those locations on the images. For example, a user may interact with a visual cue over one of those locations to modify, refine, or filter the results of a visual search, performed on a publication corpus, that uses an input image (e.g., one captured using a mobile device) as a search query.

Abstract: A biological pattern information processing device includes: a detection result obtaining unit that obtains unique region information that is detected based on biological pattern information representing a biological pattern, the unique region information representing a unique region included in the biological pattern; and a display control unit that causes the unique region to be displayed using a display attribute, based on the obtained unique region information, the display attribute being different from that for a region of the biological pattern other than the unique region.

Abstract: Methods, devices, and storage medium for fusing at least one image are disclosed. The method includes obtaining a first to-be-fused image and a second to-be-fused image, the first to-be-fused image comprising first regions, and the second to-be-fused image comprising second regions; obtaining a first feature set according to the first to-be-fused image and obtaining a second feature set according to the second to-be-fused image; performing first fusion processing on the first to-be-fused image and the second to-be-fused image by using a shape fusion network model to obtain a third to-be-fused image, the third to-be-fused image comprising at least one first encoding feature and at least one second encoding feature; and performing second fusion processing on the third to-be-fused image and the first to-be-fused image by using a condition fusion network model to obtain a target fused image. Model training methods, apparatus, and storage medium are also disclosed.

Abstract: Embodiments of the invention relate to a method of processing a medical image to remove one or more portions of the image corresponding to bone structures, the method comprising: receiving first image data representing a first, three-dimensional, medical image; processing the first image data to generate second image data representing a plurality of two-dimensional image channels each corresponding to a different slice of the first medical image; receiving the second image data at a neural network system; applying an attention mechanism at the neural network system to the second image data to generate an attention map representing one or more regions of interest; and determining, at least partly on the basis of the attention map, that one or more portions of the second image data represent a bone structure.

Abstract: A system for image reconstruction in magnetic resonance imaging (MRI) is provided. The system may obtain undersampled k-space data associated with an object, wherein the undersampled K-space data may be generated based on magnetic resonance (MR) signals collected by an MR scanner that scans the object. The system may construct an ordinary differential equation (ODE) that formulates a reconstruction of an MR image based on the undersampled k-space data. The system may further generate the MR image of the object by solving the ODE based on the undersampled k-space data using an ODE solver.

Abstract: Embodiments of a marine survey image enhancement system (MSIES) using temporal images from underwater camera are disclosed. In some embodiments, a MSIES system includes: a pre-processing module used to sub-sample input image into smaller image chunks; a DNN module constructed with many complex layers of neurons connected together and is designed to consume low quality, noisy image and produce high resolution realistic image; a post-processing module that merges all the output images from the DNN module with the original input image, forming a clearer, more detailed and higher signal to noise ratio marine survey image.