Abstract: Generation of three-dimensional (3D) object models may be challenging for users without a sufficient skill set for content creation and may also be resource intensive. One or more style transfer networks may be used for part-aware style transformation of both geometric features and textural components of a source asset to a target asset. The source asset may be segmented into particular parts and then ellipsoid approximations may be warped according to correspondence of the particular parts to the target assets. Moreover, a texture associated with the target asset may be used to warp or adjust a source texture, where the new texture can be applied to the warped parts.

Abstract: In various examples, a deep three-dimensional (3D) conditional generative model is implemented that can synthesize high resolution 3D shapes using simple guides—such as coarse voxels, point clouds, etc.—by marrying implicit and explicit 3D representations into a hybrid 3D representation. The present approach may directly optimize for the reconstructed surface, allowing for the synthesis of finer geometric details with fewer artifacts. The systems and methods described herein may use a deformable tetrahedral grid that encodes a discretized signed distance function (SDF) and a differentiable marching tetrahedral layer that converts the implicit SDF representation to an explicit surface mesh representation. This combination allows joint optimization of the surface geometry and topology as well as generation of the hierarchy of subdivisions using reconstruction and adversarial losses defined explicitly on the surface mesh.

Abstract: Generation of three-dimensional (3D) object models may be challenging for users without a sufficient skill set for content creation and may also be resource intensive. One or more style transfer networks may be combined with a generative network to generate objects based on parameters associated with a textual input. An input including a 3D mesh and texture may be provided to a trained system along with a textual input that includes parameters for object generation. Features of the input object may be identified and then tuned in accordance with the textual input to generate a modified 3D object that includes a new texture along with one or more geometric adjustments.

Abstract: Systems and methods are described for rendering complex surfaces or geometry. In at least one embodiment, neural signed distance functions (SDFs) can be used that efficiently capture multiple levels of detail (LODs), and that can be used to reconstruct multi-dimensional geometry or surfaces with high image quality. An example architecture can represent complex shapes in a compressed format with high visual fidelity, and can generalize across different geometries from a single learned example. Extremely small multi-layer perceptrons (MHLPs) can be used with an octree-based feature representation for the learned neural SDFs.

Abstract: In various examples, a deep three-dimensional (3D) conditional generative model is implemented that can synthesize high resolution 3D shapes using simple guides—such as coarse voxels, point clouds, etc.—by marrying implicit and explicit 3D representations into a hybrid 3D representation. The present approach may directly optimize for the reconstructed surface, allowing for the synthesis of finer geometric details with fewer artifacts. The systems and methods described herein may use a deformable tetrahedral grid that encodes a discretized signed distance function (SDF) and a differentiable marching tetrahedral layer that converts the implicit SDF representation to an explicit surface mesh representation. This combination allows joint optimization of the surface geometry and topology as well as generation of the hierarchy of subdivisions using reconstruction and adversarial losses defined explicitly on the surface mesh.

Abstract: Systems and methods are described for rendering complex surfaces or geometry. In at least one embodiment, neural signed distance functions (SDFs) can be used that efficiently capture multiple levels of detail (LODs), and that can be used to reconstruct multi-dimensional geometry or surfaces with high image quality. An example architecture can represent complex shapes in a compressed format with high visual fidelity, and can generalize across different geometries from a single learned example. Extremely small multi-layer perceptrons (MLPs) can be used with an octree-based feature representation for the learned neural SDFs.

Abstract: Approaches presented herein can utilize a network that learns to embed three-dimensional (3D) coordinates on a surface of one or more 3D shapes into an aligned two-dimensional (2D) texture space, where corresponding parts of different 3D shapes can be mapped to the same location in a texture image. Alignment can be performed using a texture alignment module that generates a set of basis images for synthesizing textures. A trained network can generate a basis shared by all shape textures, and can predict input-specific coefficients to construct the output texture for each shape as a linear combination of the basis images, then deform the texture to match the pose of the input. Such an approach can ensure alignment of textures, even in situations with at least somewhat limited network capacity. To unwrap shapes of complex structure or topology, a masking network can be utilized that cuts the shape into multiple pieces to reduce the distortion in the 2D mapping.

Abstract: Generation of three-dimensional (3D) object models may be challenging for users without a sufficient skill set for content creation and may also be resource intensive. One or more style transfer networks may be used for part-aware style transformation of both geometric features and textural components of a source asset to a target asset. The source asset may be segmented into particular parts and then ellipsoid approximations may be warped according to correspondence of the particular parts to the target assets. Moreover, a texture associated with the target asset may be used to warp or adjust a source texture, where the new texture can be applied to the warped parts.

Abstract: In various examples, a deep three-dimensional (3D) conditional generative model is implemented that can synthesize high resolution 3D shapes using simple guides—such as coarse voxels, point clouds, etc.—by marrying implicit and explicit 3D representations into a hybrid 3D representation. The present approach may directly optimize for the reconstructed surface, allowing for the synthesis of finer geometric details with fewer artifacts. The systems and methods described herein may use a deformable tetrahedral grid that encodes a discretized signed distance function (SDF) and a differentiable marching tetrahedral layer that converts the implicit SDF representation to an explicit surface mesh representation. This combination allows joint optimization of the surface geometry and topology as well as generation of the hierarchy of subdivisions using reconstruction and adversarial losses defined explicitly on the surface mesh.

Abstract: Systems and methods are described for rendering complex surfaces or geometry. In at least one embodiment, neural signed distance functions (SDFs) can be used that efficiently capture multiple levels of detail (LODs), and that can be used to reconstruct multi-dimensional geometry or surfaces with high image quality. An example architecture can represent complex shapes in a compressed format with high visual fidelity, and can generalize across different geometries from a single learned example. Extremely small multi-layer perceptrons (MLPs) can be used with an octree-based feature representation for the learned neural SDFs.

Abstract: Systems and methods are described for rendering complex surfaces or geometry. In at least one embodiment, neural signed distance functions (SDFs) can be used that efficiently capture multiple levels of detail (LODs), and that can be used to reconstruct multi-dimensional geometry or surfaces with high image quality. An example architecture can represent complex shapes in a compressed format with high visual fidelity, and can generalize across different geometries from a single learned example. Extremely small multi-layer perceptrons (MLPs) can be used with an octree-based feature representation for the learned neural SDFs.

Abstract: Systems and methods are described for rendering complex surfaces or geometry. In at least one embodiment, neural signed distance functions (SDFs) can be used that efficiently capture multiple levels of detail (LODs), and that can be used to reconstruct multi-dimensional geometry or surfaces with high image quality. An example architecture can represent complex shapes in a compressed format with high visual fidelity, and can generalize across different geometries from a single learned example. Extremely small multi-layer perceptrons (MLPs) can be used with an octree-based feature representation for the learned neural SDFs.

Abstract: The disclosure relates processing point cloud data based on a neural network. The method includes: acquiring first point set data and second point set data respectively representing a point data set of an outer surface of a target object in different shapes; processing the first point set data using structure parameters of the neural network to obtain a multi-scale feature of each point in the first point set data; acquiring target point set data according to the first displacement vector, the first point set data and the second point set data. The method of processing the point cloud data provided in the present application can implement shape conversion of the point cloud.

Abstract: The disclosure relates processing point cloud data based on a neural network. The method includes: acquiring first point set data and second point set data respectively representing a point data set of an outer surface of a target object in different shapes; processing the first point set data using structure parameters of the neural network to obtain a multi-scale feature of each point in the first point set data; acquiring target point set data according to the first displacement vector, the first point set data and the second point set data. The method of processing the point cloud data provided in the present application can implement shape conversion of the point cloud.

Abstract: A method for reconstructing surface from a point cloud includes following steps: (a) extracting skeletal curves from an input point cloud; (b) editing the extracted skeletal curves, and assigning sweeping path; (c) obtaining sliced point clouds along the edited skeletal curves, and fitting a closed NURBS curve according to the sliced point clouds; (d) reconstructing the point cloud to get generalized cylinders along the assigned sweeping path, according to the closed NURBS curves; (e) merging the generalized cylinders into a single surface, and smoothing intersections of the generalized cylinders so as to reconstruct surface from the point cloud. The invention further relates to a system for reconstructing surface from a point cloud. The invention can reconstruct the surface with high accuracy by the minimum interactions, and can deal with point cloud data having missing region caused by occlusion. In addition, the invention can achieve high reconstruction quality and fine controllability.

Abstract: Provided is a method for extracting an image salient curve. The method comprises the following steps: drawing an approximate curve along a salient edge of an image from which a salient curve is to be extracted; obtaining short edges in the image; calculating a harmonic vector field by using the drawn curve as a boundary condition; filtering the short edges in the image by using the harmonic vector field; updating the vector field by using the short edges left in the image as boundary conditions; and obtaining an optimal salient curve of the image by using the energy of a minimized spline curve in the vector field. Also provided is a system for extracting an image salient curve. The image salient curve can ensure the smoothness and a bending characteristic.

Abstract: A method for reconstructing surface from a point cloud includes following steps: (a) extracting skeletal curves from an input point cloud; (b) editing the extracted skeletal curves, and assigning sweeping path; (c) obtaining sliced point clouds along the edited skeletal curves, and fitting a closed NURBS curve according to the sliced point clouds; (d) reconstructing the point cloud to get generalized cylinders along the assigned sweeping path, according to the closed NURBS curves; (e) merging the generalized cylinders into a single surface, and smoothing intersections of the generalized cylinders so as to reconstruct surface from the point cloud. The invention further relates to a system for reconstructing surface from a point cloud. The invention can reconstruct the surface with high accuracy by the minimum interactions, and can deal with point cloud data having missing region caused by occlusion. In addition, the invention can achieve high reconstruction quality and fine controllability.

Abstract: Provided are an image restoration method and device. The method comprises: processing image blocks, which are initially registered, to acquire connection curves among the image blocks; constructing an ambient field of images to be restored by means of the connection curve; by minimizing energy of the connection curve in the ambient field, registering the image blocks; and performing image filling on the registered image blocks to acquire a restored image. The device comprises: a processing unit used for processing the image blocks, which are initially registered, to acquire the connection curve among the image blocks; an ambient field construction unit used for constructing the ambient field of the image to be restored by means of the connection curve; a registering unit used for registering the image blocks by minimizing the energy of the connection curve in the ambient field; and a filling unit used for performing image filling on the registered image blocks to acquire the restored image.

Abstract: Provided is a method for extracting an image salient curve. The method comprises the following steps: drawing an approximate curve along a salient edge of an image from which a salient curve is to be extracted; obtaining short edges in the image; calculating a harmonic vector field by using the drawn curve as a boundary condition; filtering the short edges in the image by using the harmonic vector field; updating the vector field by using the short edges left in the image as boundary conditions; and obtaining an optimal salient curve of the image by using the energy of a minimized spline curve in the vector field. Also provided is a system for extracting an image salient curve. The image salient curve can ensure the smoothness and a bending characteristic.

Abstract: Provided are an image restoration method and device. The method comprises: processing image blocks, which are initially registered, to acquire connection curves among the image blocks; constructing an ambient field of images to be restored by means of the connection curve; by minimizing energy of the connection curve in the ambient field, registering the image blocks; and performing image filling on the registered image blocks to acquire a restored image. The device comprises: a processing unit used for processing the image blocks, which are initially registered, to acquire the connection curve among the image blocks; an ambient field construction unit used for constructing the ambient field of the image to be restored by means of the connection curve; a registering unit used for registering the image blocks by minimizing the energy of the connection curve in the ambient field; and a filling unit used for performing image filling on the registered image blocks to acquire the restored image.