CN-120672956-B - Four-sided heavy topology method based on frame field prediction and electronic equipment

CN120672956BCN 120672956 BCN120672956 BCN 120672956BCN-120672956-B

Abstract

The invention provides a frame field prediction-based four-sided heavy topology method and electronic equipment. The four-sided heavy topology method based on the frame field prediction comprises the steps of inputting a triangular grid model into a frame field neural network model to conduct direction regression and amplitude diffusion taking the direction as a condition to obtain a corresponding frame field, deforming based on the frame field to obtain a deformed grid model, conducting isotropic quadrilateral re-grid division on the deformed grid model to obtain an initial quadrilateral grid model, and conducting inverse deformation on the initial quadrilateral grid model to obtain a quadrilateral grid model corresponding to the triangular grid model. The quality of the finally generated quadrilateral mesh is ensured through the steps of frame field inference, deformation, isotropic quadrilateral re-mesh division and inversion, so that the aim of improving the quality of the quadrilateral mesh is fulfilled.

Inventors

LIU YINGTIAN
GUO YUANCHEN
CAO YANPEI
LIANG DING

Assignees

北京哇嘶嗒科技有限公司

Dates

Publication Date: 20260505
Application Date: 20250611

Claims (7)

1. A frame field prediction-based four-sided heavy topology method, comprising: Inputting the triangular mesh model into a framework field neural network model for direction regression and amplitude diffusion taking the direction as a condition to obtain a corresponding framework field; Deforming based on the frame field to obtain a deformed grid model; performing isotropic quadrilateral re-grid division on the deformed grid model to obtain an initial quadrilateral grid model; performing inverse deformation on the initial quadrilateral mesh model to obtain a quadrilateral mesh model corresponding to the triangular mesh model; taking a frame field as a point attribute in the training of the neural network model of the frame field, using point coding geometric information with normal, and using multi-vector representation frame; The direction prediction of the frame field neural network model includes, for a point cloud acquired from triangle mesh samples, expressing true values of a predicted direction vector d using a loss function The loss function formula is: ; the frame field neural network model uses diffusion for amplitude prediction, the using diffusion for amplitude prediction comprising: Using log-amplitude as a diffusion target; In each training step, time step T is uniformly sampled in { 1..once, T } and the forward diffusion process is , wherein, Representing the amplitude, epsilon is the random noise extracted from the standard normal distribution, , Is a pre-calculated constant associated with the noise scheduler.
2. The frame field prediction based four-sided re-topology method of claim 1, wherein inputting a triangular mesh model into a frame field neural network model comprises: Isotropically re-meshing the triangular mesh model to obtain a partitioned triangular mesh model; Uniformly sampling the divided triangular mesh model to obtain sampling points; The center point and the sampling point of each triangle mesh are input into a frame field neural network model.
3. The frame field prediction based four-sided heavy topology method of claim 1, wherein the deforming based on the frame field results in a deformed mesh model, comprising: Decomposing each frame of the frame field into a product of a linear SPD map and a cross product; The ARAP energy function is minimized, and the position of the vertex after deformation is solved; and obtaining the deformed grid model based on the deformed vertex positions.
4. The frame field prediction based four-sided heavy topology method of claim 1, wherein the isotropically quadrilateral re-meshing of the deformed mesh model to obtain an initial quadrilateral mesh model comprises: Isotropic quadrilateral re-meshing of the deformed mesh model is performed using an isotropic quadrilateral re-meshing algorithm with cross-product field constraints.
5. The frame field prediction based four-sided re-topology method of claim 4, wherein isotropically quad-rescheduling the warped mesh model using an isotropically quad-rescheduling algorithm with cross product field constraints, comprising: re-meshing the deformed mesh model into an initial quadrilateral mesh model by using QuadriFlow; Using QuadriFlow, the cross-product at the vertex is calculated as the sum of the cross-products on the faces, weighted by the area of the adjacent faces, after the local alignment.
6. The frame field prediction based four-sided heavy topology method of claim 1, wherein said inverse deforming the initial quadrilateral mesh model comprises: And (3) carrying out inverse deformation on the initial quadrilateral mesh model by adopting a deformation transfer algorithm, wherein the deformed mesh model in the deformation transfer algorithm is regarded as a source mesh, the initial quadrilateral mesh model is regarded as a target mesh, and the triangular mesh model is regarded as a gesture source mesh.
7. An electronic device, the electronic device comprising: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the frame field prediction based four-sided heavy topology method of any of claims 1-6.

Description

Four-sided heavy topology method based on frame field prediction and electronic equipment Technical Field The invention belongs to the technical field of three-dimensional models, and particularly relates to a frame field prediction-based four-sided heavy topology method and electronic equipment. Background Quadrilateral grids play a critical role in computer graphics applications, however, automatically generating high quality quadrilateral grids remains challenging. With the explosive development of modeling techniques and 3D content representations, 3D model generation has received great attention from communities over the past few years. In particular, 3D models defined with quadrilateral meshes play an important role in computer aided design, physical simulation, and character animation. However, designing a high quality quadrilateral mesh is labor intensive and requires domain specific knowledge. Therefore, it is of great value to develop a method for automatically generating a high quality quadrilateral mesh from intuitive user input. Contemporary 3D content generation and reconstruction methods have met with significant success by employing neural fields as their underlying geometric representations. Such a representation naturally adapts to the generation method and enables a high fidelity shape synthesis. While these methods typically use established field discretization techniques to extract triangular meshes from implicit surfaces, the resulting triangulated geometry typically lacks the geometric quality and structural regularity inherent in quadrilateral dominant meshes created by professional artists. This quality difference, particularly in terms of mesh topology and element alignment, has prompted the development of powerful quadrilateral techniques to bridge the gap between neural field reconstruction and the assets available for production. The prior art has the following problems when generating quadrilateral meshes: 1. Conventional quadrilateral mesh generation methods rely primarily on local geometric features or manually specified constraints to guide the alignment of mesh elements. This results in suboptimal generated quadrilateral layouts in areas where local disparity attributes provide insufficient guidance (e.g., near flat areas or areas with complex curvature patterns), failing to capture global shape semantics. 2. Non-manifold meshes cannot be handled efficiently-existing mesh-based neural architectures have difficulty handling non-manifold meshes, which account for a significant proportion of real-world 3D content. 3. The lack of large-scale high quality data sets early learning-based approaches had limited training data due to the lack of large-scale high quality data sets, which may hamper their generalization ability. Disclosure of Invention Aiming at the problems in the prior art, the invention provides a frame field prediction-based quadrilateral surface heavy topology method and electronic equipment, and at least partially solves the problem of low quality of quadrilateral grids in the prior art. In a first aspect, an embodiment of the present disclosure provides a method for four-sided planar heavy topology based on frame field prediction, including: Inputting the triangular mesh model into a framework field neural network model for direction regression and amplitude diffusion taking the direction as a condition to obtain a corresponding framework field; Deforming based on the frame field to obtain a deformed grid model; performing isotropic quadrilateral re-grid division on the deformed grid model to obtain an initial quadrilateral grid model; and carrying out inverse deformation on the initial quadrilateral mesh model so as to obtain a quadrilateral mesh model corresponding to the triangular mesh model. Optionally, inputting the triangular mesh model into the frame field neural network model includes: Isotropically re-meshing the triangular mesh model to obtain a partitioned triangular mesh model; Uniformly sampling the divided triangular mesh model to obtain sampling points; The center point and the sampling point of each triangle mesh are input into a frame field neural network model. Optionally, the deforming based on the frame field to obtain a deformed grid model includes: Decomposing each frame of the frame field into a product of a linear SPD map and a cross product; The ARAP energy function is minimized, and the position of the vertex after deformation is solved; and obtaining the deformed grid model based on the deformed vertex positions. Optionally, the performing isotropic quadrilateral re-meshing on the deformed mesh model to obtain an initial quadrilateral mesh model includes: Isotropic quadrilateral re-meshing of the deformed mesh model is performed using an isotropic quadrilateral re-meshing algorithm with cross-product field constraints. Optionally, the isotropic quadrilateral re-meshing of the deformed mesh model using an isotropic quadrilateral re-mes