Abstract: A three-dimensional point cloud upsampling method includes: dividing a three-dimensional point cloud into overlappable point cloud blocks which have a fixed number of points and are capable of covering all the points; extracting hierarchical features according to point coordinates in the point cloud blocks; achieving point set feature expansion of the extracted hierarchical features by using multi-scale heat kernel graph convolution; and reconstructing point coordinates in an upsampled three-dimensional point cloud from the expanded features. According to the present disclosure, detail information enhancement with different fine granularities can be performed on the three-dimensional point cloud which is sparse and nonuniformly distributed in the space, and at the same time, good stability is provided for overcoming potential noise disturbance and local deformation.

Abstract: The present disclosure provides an image classification method for maximizing mutual information, device, medium and system, the method including: acquiring a training image; maximizing the mutual information between the training image and a neural network architecture, and automatically determining the network architecture and parameter of the neural network; and processing image data to be classified using the obtained neural network to obtain an image classification result. According to the present disclosure, the network architecture and parameter of the neutral network are automatically designed and determined by maximizing the mutual information based on given image data without burdensome manual design and saving human and computational resource consumption. The present disclosure can automatically design and obtain a neural network-based image classification method in a very short time, and at the same time can achieve higher image classification accuracy.

Abstract: A graph dictionary learning method for a 3D point cloud comprises: obtaining N point clouds to form training dataset; performing voxelization process on the point cloud data to obtain voxelized point cloud data of the training dataset; performing voxel block division on the point cloud data of the training dataset, selecting a plurality of voxel blocks as the training dataset, and constructing a graph dictionary learning model according to the training dataset; and performing iterative optimization on the graph dictionary learning objective function to obtain a graph dictionary for encoding and decoding a 3D point cloud signal. The present disclosure effectively uses the spatial correlation between point cloud signals to near-optimally remove the redundancy among point cloud signals.

Abstract: The present disclosure provides an image classification method for maximizing mutual information, device, medium and system, the method including: acquiring a training image; maximizing the mutual information between the training image and a neural network architecture, and automatically determining the network architecture and parameter of the neural network; and processing image data to be classified using the obtained neural network to obtain an image classification result. According to the present disclosure, the network architecture and parameter of the neutral network are automatically designed and determined by maximizing the mutual information based on given image data without burdensome manual design and saving human and computational resource consumption. The present disclosure can automatically design and obtain a neural network-based image classification method in a very short time, and at the same time can achieve higher image classification accuracy.

Abstract: A graph dictionary learning method for a 3D point cloud comprises: obtaining N point clouds to form training dataset; performing voxelization process on the point cloud data to obtain voxelized point cloud data of the training dataset; performing voxel block division on the point cloud data of the training dataset, selecting a plurality of voxel blocks as the training dataset, and constructing a graph dictionary learning model according to the training dataset; and performing iterative optimization on the graph dictionary learning objective function to obtain a graph dictionary for encoding and decoding a 3D point cloud signal. The present disclosure effectively uses the spatial correlation between point cloud signals to near-optimally remove the redundancy among point cloud signals.

Abstract: A three-dimensional point cloud upsampling method includes: dividing a three-dimensional point cloud into overlappable point cloud blocks which have a fixed number of points and are capable of covering all the points; extracting hierarchical features according to point coordinates in the point cloud blocks; achieving point set feature expansion of the extracted hierarchical features by using multi-scale heat kernel graph convolution; and reconstructing point coordinates in an upsampled three-dimensional point cloud from the expanded features. According to the present disclosure, detail information enhancement with different fine granularities can be performed on the three-dimensional point cloud which is sparse and nonuniformly distributed in the space, and at the same time, good stability is provided for overcoming potential noise disturbance and local deformation.

Abstract: An image processing method for sparse image reconstruction, image denoising, compressed sensing image reconstruction or image restoration, comprising: establishing a general linear optimization inverse problem under the 1-norm constraint of a sparse signal; establishing a differentiable deep network model based on convex combination to solve the problem on the basis of standard or learned iterative soft shrinkage thresholding algorithm; and introducing a deep neural network of arbitrary structure into the solving step to accelerate the solving step and reducing a number of iterations needed to reach a convergence. The present disclosure combines the traditional iterative optimization algorithm with the deep neural network of arbitrary structure to improve the image reconstruction performance and ensure fast convergence to meet the current needs of sparse image reconstruction.

Abstract: An image processing method for sparse image reconstruction, image denoising, compressed sensing image reconstruction or image restoration, comprising: establishing a general linear optimization inverse problem under the 1-norm constraint of a sparse signal; establishing a differentiable deep network model based on convex combination to solve the problem on the basis of standard or learned iterative soft shrinkage thresholding algorithm; and introducing a deep neural network of arbitrary structure into the solving step to accelerate the solving step and reducing a number of iterations needed to reach a convergence. The present disclosure combines the traditional iterative optimization algorithm with the deep neural network of arbitrary structure to improve the image reconstruction performance and ensure fast convergence to meet the current needs of sparse image reconstruction.

Abstract: In a 3D point cloud compression system based on multi-scale structured dictionary learning, a point cloud data partition module outputs a voxel set and a set of blocks of voxels of different scales. A geometric information encoding module outputs an encoded geometric information bit stream. A geometric information decoding module outputs decoded geometric information. An attribute signal encoding module outputs a sparse coding coefficient matrix and a learned multi-scale structured dictionary. An attribute signal compression module outputs a compressed attribute signal bit stream. An attribute signal decoding module outputs decoded attribute signals. A 3D point cloud reconstruction module completes reconstruction. The system is applicable to lossless geometric and lossy attribute compression of point cloud signals.

Abstract: The present disclosure provides a saliency prediction method and system for a 360-degree image based on a graph convolutional neural network. The method includes: firstly, constructing a spherical graph signal of an image of an equidistant rectangular projection format by using a geodesic icosahedron composition method; then inputting the spherical graph signal into the proposed graph convolutional neural network for feature extraction and generation of a spherical saliency graph signal; and then reconstructing the spherical saliency graph signal into a saliency map of an equidistant rectangular projection format by using a proposed spherical crown based interpolation algorithm. The present disclosure further proposes a KL divergence loss function with sparse consistency. The method can achieve excellent saliency prediction performance subjectively and objectively, and is superior to an existing method in computational complexity.

Abstract: The present disclosure provides a saliency prediction method and system for a 360-degree image based on a graph convolutional neural network. The method includes: firstly, constructing a spherical graph signal of an image of an equidistant rectangular projection format by using a geodesic icosahedron composition method; then inputting the spherical graph signal into the proposed graph convolutional neural network for feature extraction and generation of a spherical saliency graph signal; and then reconstructing the spherical saliency graph signal into a saliency map of an equidistant rectangular projection format by using a proposed spherical crown based interpolation algorithm. The present disclosure further proposes a KL divergence loss function with sparse consistency. The method can achieve excellent saliency prediction performance subjectively and objectively, and is superior to an existing method in computational complexity.

Abstract: In a 3D point cloud compression system based on multi-scale structured dictionary learning, a point cloud data partition module outputs a voxel set and a set of blocks of voxels of different scales. A geometric information encoding module outputs an encoded geometric information bit stream. A geometric information decoding module outputs decoded geometric information. An attribute signal encoding module outputs a sparse coding coefficient matrix and a learned multi-scale structured dictionary. An attribute signal compression module outputs a compressed attribute signal bit stream. An attribute signal decoding module outputs decoded attribute signals. A 3D point cloud reconstruction module completes reconstruction. The system is applicable to lossless geometric and lossy attribute compression of point cloud signals.

Abstract: A hair styling device is described that has a handle, an elongated, substantially cylindrical, heatable rod providing a curling surface and supported by the handle, a heater for heating the curling surface, a support spaced apart from the curling surface, and a rotatable member rotatable relative to the curling surface for wrapping hair around and in contact with the curling surface. The device has an opening in the rotatable member for receiving a section of a user's hair, a motor for rotating the rotatable member, and a flexible, heat-resistant clamp supported by the support and extending from the support toward the curling surface for clamping hair against the curling surface.

Abstract: A hair styling device is described that has a handle, an elongated, substantially cylindrical, heatable rod providing a curling surface and supported by the handle, a heater for heating the curling surface, a support spaced apart from the curling surface, and a rotatable member rotatable relative to the curling surface for wrapping hair around and in contact with the curling surface. The device has an opening in the rotatable member for receiving a section of a user's hair, a motor for rotating the rotatable member, and a flexible, heat-resistant clamp supported by the support and extending from the support toward the curling surface for clamping hair against the curling surface.

Abstract: A hair styling device is described that has a handle, an elongated, substantially cylindrical, heatable rod providing a curling surface and supported by the handle, a heater for heating the curling surface, a support spaced apart from the curling surface, and a rotatable member rotatable relative to the curling surface for wrapping hair around and in contact with the curling surface. The device has an opening in the rotatable member for receiving a section of a user's hair, a motor for rotating the rotatable member, and a flexible, heat-resistant clamp supported by the support and extending from the support toward the curling surface for clamping hair against the curling surface.