Abstract: An image depth prediction method acquires image frames of containing a dynamic object by a monocular camera, extracts a continuous of object frames and reference frames from the image frames, reconstructs the object frames to obtain reconstructed frames according to the reference frames and a preset depth estimation model, obtains a reconstruction error between the object frames and the reconstructed frames, processes the image frames to obtain point cloud data and instance segmentation data, fuses the point cloud data with the instance segmentation data to obtain mask data, obtains a loss function according to the reconstruction error and the mask data, and trains the depth estimation model based on the loss function until the loss function converges. The method can obtain more accurate depth estimation results for dynamic scenes. An electronic device and a non-transitory storage recording the method are also disclosed.

Abstract: An image analysis method includes the steps of obtaining a saturation channel image and a value channel image according to an original image; obtaining a relative deviation value image based on the saturation channel image and the value channel image, wherein the relative deviation value image represents relative deviations between saturation and value; extracting several contours from the relative deviation value image; determining target parameter values corresponding to each contour; and Determining according to analysis results of the target parameter values corresponding to each contour whether there is a non-working region in the original image. A related image analysis apparatus, computer device, and computer readable storage medium are also disclosed.

Abstract: A dynamic and selective tone mapping system that receives digital imagery content and adjusts the digital imagery content using defined perceptual point parameters that correspond to particular luminance points or regions of the digital imagery content. The defined perceptual point parameters include a first white reference value, wherein the tone mapping system adjusts the digital imagery content by compressing luminance values above the first white reference value at different compression rates to produce digital imagery content that is compatible with various media viewing devices with varying display brightness capabilities while preserving influence over the intended look of the displayed image resulting from the digital imagery content.

Abstract: Described herein are methods, systems and computer-program products for intra- and inter-gait classification for both symmetric and asymmetric gait patterns using images.

Abstract: Aspects of the subject disclosure relate to stereo vision-based height clearance detection, of which a device includes a processor that obtains first data and second data from different cameras of a vehicle. The processor generates a depth map based on a disparity between the first data and the second data. The processor determines a road segmentation based at least in part on the depth map. The processor determines road surface height information for corresponding portions of the road segmentation. The processor determines a height estimation of a non-road object corresponding to at least one portion of the road segmentation. The processor provides an indication of a height clearance estimate for the at least one portion of the road segmentation based on the road surface height information and the height estimation of the non-road object.

Abstract: A control method, device, equipment and storage medium for interactive reproduction of a target object is provided. The control method comprises: acquiring a control instruction of the target object, the control instruction including a plurality of direction control parameters; acquiring target image information of the target object according to the combination of the plurality of direction control parameters, the target image information being acquired through a preset three-dimensional reproduction method; performing interactive reproduction on the target object based on the direction control parameters and the target image information.

Abstract: Techniques for spatially preserving flattening in deep learning neural networks are provided. In one aspect, a spatially preserving flattening module includes: a predictor for generating image feature maps from at least one convolutional layer of a feature extraction phase of a deep learning neural network applied to input image data; an auto-encoder for producing encodings of the image feature maps that preserve location and shape information associated with objects in the input image data; and a flattener for concatenating the encodings of the image feature maps to form a spatially preserving flattened encoding vector. A deep learning neural network that includes the present spatially preserving flattening module is also provided, as is a method for spatially preserving flattening.

Abstract: An image registration method and apparatus, an electronic apparatus, and a storage medium are provided. The image registration method comprises: extracting edge pixels of the binocular image, and determining high-confidence parallax points in the edge pixels and parallax; projecting each non-high-confidence parallax point to the triangular mesh in a direction parallel with a parallax dimension to obtain a triangular face; determining a parallax search range of parallax of each non-high-confidence parallax point, calculating a matching cost corresponding to all parallax in each parallax search range, and determining that parallax with the smallest matching cost is the parallax of the corresponding non-high-confidence parallax point; and determining a depth boundary point in the edge pixels, determining a parallax boundary point of each depth boundary point in a target direction, and setting parallax of pixels between the depth boundary point and the parallax boundary point to a target value.

Abstract: Methods and apparatus are disclosed herein to synthesize six degree-of-freedom views from sparce RGB-depth inputs. An example apparatus includes at least one memory, instructions in the apparatus, and processor circuitry to execute the instructions to reproject a first red-green-blue (RGB) depth view and a second RGB depth view to a target camera position to obtain a first reprojected view and a second reprojected view, combine the first reprojected view and the second reprojected view into a blended view data based on a first weight map and a second weight map, the blended view including missing RGB depth information due to at least one of an occlusion or a disocclusion, and generate a six degree-of-freedom synthesized view of the blended view data, the synthesized view including the missing RGB depth information.

Abstract: A neural network training method, includes: obtaining an input feature map of a training image; performing feature extraction processing on the input feature map by using a feature extraction core of a neural network to obtain a first candidate feature map; adding the first candidate feature map and a second candidate feature map to obtain an output feature map, where the second candidate feature map is a feature map obtained after a value corresponding to each element in the input feature map is increased by N times, and N is greater than 0; determining an image processing result of the training image based on the output feature map; and adjusting a parameter of the neural network based on the image processing result.

Abstract: An object is to obtain three-dimensional shape data with high accuracy from three-dimensional shape data representing an approximate shape of an object. Three-dimensional shape data of an object captured in a plurality of captured images whose viewpoints are different is obtained. Further, surface three-dimensional information on the object is derived based on the plurality of captured images. Then, the derived surface three-dimensional information is selected based on a distance from the shape surface of the object represented by the three-dimensional shape data.

Abstract: A fully automated and reliable picking of a diverse range of unseen objects in clutter is a challenging problem. The present disclosure provides an optimum grasp pose selection to pick an object from a bin. Initially, the system receives an input image pertaining to a surface. Further, a plurality of sampled grasp poses are generated in a random configuration. Further, a depth difference value is computed for each of a plurality of pixels corresponding to each of the plurality of sampled grasp poses. Further, a binary map is generated for each of the plurality of sampled grasp poses and a plurality of subregions are obtained. Further, a plurality of feasible grasp poses are selected based on the plurality of subregions and a plurality of conditions. Further, the plurality of feasible grasp poses are refined and an optimum grasp pose is obtained based on a Grasp Quality Score (GQS).

Abstract: A neural light field (NeLF) that runs real-time on mobile devices for neural rendering of three dimensional (3D) scenes, referred to as MobileR2L. The MobileR2L architecture runs efficiently on mobile devices with low latency and small size, and it achieves high-resolution generation while maintaining real-time inference for both synthetic and real-world 3D scenes on mobile devices. The MobileR2L has a network backbone including a convolutional layer embedding an input image at a resolution, residual blocks uploading the embedded image, and super-resolution modules receiving the uploaded embedded image and rendering an output image having a higher resolution than the embedded image. The convolution layer generates a number of rays equal to a number of pixels in the input image, where a partial number of the rays is uploaded to the super-resolution modules.

Abstract: An EDOF signal processing unit performs EDOF signal processing as restoration processing on a special light image acquired by imaging an observation target irradiated with special light, and a setting unit sets a restoration degree of the EDOF signal processing for the EDOF signal processing unit. Then, a parameter for the restoration degree in the EDOF signal processing on the special light image is set so as to make the restoration degree in the EDOF signal processing on the special light image lower than a restoration degree in EDOF signal processing as restoration processing on a normal light image acquired by imaging the observation target irradiated with normal light. The present technology may be applied to, for example, a medical image processing system that performs special light observation.

Abstract: The driver monitoring device includes a processor configured to detect an opening/closing state of eyes of a driver of a vehicle based on a facial image in which a face of the driver is shown, judge whether the driver has fallen asleep based on the opening/closing state of the eyes, and judge whether the driver is talking. The processor is configured not to judge that the driver has fallen asleep while it is judged that the driver is talking.

Abstract: The present disclosure relates to an image registration method and a model training method thereof. The image registration method comprises obtaining a reference image and a floating image to be registered, performing image preprocessing on the reference image and the floating image, performing non-rigid registration on the preprocessed reference image and floating image to obtain a registration result image, and outputting the registration result image. The image preprocessing comprises performing, on the reference image and the floating image, coarse-to-fine rigid registration based on iterative closest point registration and mutual information registration. The non-rigid registration uses a combination of a correlation coefficient and a mean squared error between the reference image and the registration result image as a loss function.

Abstract: The invention relates to the technical field of intelligent construction and maintenance, and relates to a non-contact visual monitoring system and method for a flexible protective structure against rockfall disasters; it comprises a hardware system and a data analysis system, and the hardware system uses a multipoint distributed high-resolution high-speed camera to capture dynamic image sequences of a protective structure under the rockfall impact in a non-contact mode; the data analysis system comprises an impact deformation state full-field tracking module and a multipoint impact large deformation extraction module, and the impact deformation state full-field tracking module captures spatio-temporal changes of impact deformation of the protective system by adopting a full-field optical flow method, constructs a two-dimensional velocity amplitude distribution diagram and performs full-field spatio-temporal tracking of large deformation of the system.

Abstract: One or more processing devices access a scene depicting a reference object that includes an annotation identifying a target region to be modified in one or more video frames. The one or more processing devices determine that a target pixel corresponds to a sub-region within the target region that includes hallucinated content. The one or more processing devices determine gradient constraints using gradient values of neighboring pixels in the hallucinated content, the neighboring pixels being adjacent to the target pixel and corresponding to four cardinal directions. The one or more processing devices update color data of the target pixel subject to the determined gradient constraints.

Abstract: Disclosed frameworks for generating an image including a salient object and a staged background include extracting a salient object from a source image and applying a generative model to the salient object to generate the image. According to some embodiments, extracting a salient object from a source image involves using salient object detection method to identify the relevant portions of the source image corresponding to the salient object. In some embodiments, the generative model is a generative adversarial network trained using a domain relevant dataset.

Abstract: This disclosure describes entropy coding techniques for media data coded using neural-based techniques. A media coder is configured to determine a probability distribution function parameter for a data element of a data stream coded by a neural-based media compression technique, wherein the probability distribution function parameter is a logarithmic function of a standard deviation of a probability distribution function of the data stream, determine a code vector based on the probability distribution function parameter, and entropy code the data element using the code vector.