“…Furthermore, most of our experimental results show that there are more than 2 meaningful components extracted from the second level AEs. So we choose Kz = [10,5]. For the qualitative evaluation, we show the results in Fig.…”
Section: The Choice Of K Zmentioning
confidence: 99%
“…As shown in that section, our stacked AEs has the lowest error with the following default parameters: λ 1 = 10.0, λ 2 = 1.0, where λ 1 , λ 2 are the weights of reconstruction error and sparsity constrain term, respectively. d = [d 1 , d 2 ] = [0.4, 0.2], K z = [10,5], corresponding to the coarse and fine levels. Here, we train the whole network end-to-end rather than in a separate manner.…”
Section: Implementation Detailsmentioning
confidence: 99%
“…So we visualize each extracted components in each level. Due to our setting (K z = [10,5]), the first level and second level autoencoders output 10 and 5 deformation components respectively. For all results, there are some slight deformation components that is not shown for the brief and reasonable visualization.…”
Section: Qualitative Evaluationmentioning
confidence: 99%
“…Since methods compared in the paper produce 50 deformation components, we let K z0 × K z1 = 50. We have the following combinations: K z = [1,50], [2,25], [5,10], [10,5], [25,2], [50,1]. However, the [1,50], [50,1] are the trivial setting.…”
Section: The Choice Of K Zmentioning
confidence: 99%
“…Multi-scale techniques are getting increasingly popular in various fields. In Finite Element Methods, multiscale analysis is widely used [11], [12], [13]. In the spectral geometry field, research works apply multiscale technology on the deformation representation [14], physics-based simulation of deformable objects [15], and surface registration [16] by analyzing the nonisometric global and local (multiscale) deformation.…”
Deformation component analysis is a fundamental problem in geometry processing and shape understanding. Existing approaches mainly extract deformation components in local regions at a similar scale while deformations of real-world objects are usually distributed in a multi-scale manner. In this paper, we propose a novel method to exact multiscale deformation components automatically with a stacked attention-based autoencoder. The attention mechanism is designed to learn to softly weight multi-scale deformation components in active deformation regions, and the stacked attention-based autoencoder is learned to represent the deformation components at different scales. Quantitative and qualitative evaluations show that our method outperforms state-of-the-art methods. Furthermore, with the multiscale deformation components extracted by our method, the user can edit shapes in a coarse-to-fine fashion which facilitates effective modeling of new shapes.
“…Furthermore, most of our experimental results show that there are more than 2 meaningful components extracted from the second level AEs. So we choose Kz = [10,5]. For the qualitative evaluation, we show the results in Fig.…”
Section: The Choice Of K Zmentioning
confidence: 99%
“…As shown in that section, our stacked AEs has the lowest error with the following default parameters: λ 1 = 10.0, λ 2 = 1.0, where λ 1 , λ 2 are the weights of reconstruction error and sparsity constrain term, respectively. d = [d 1 , d 2 ] = [0.4, 0.2], K z = [10,5], corresponding to the coarse and fine levels. Here, we train the whole network end-to-end rather than in a separate manner.…”
Section: Implementation Detailsmentioning
confidence: 99%
“…So we visualize each extracted components in each level. Due to our setting (K z = [10,5]), the first level and second level autoencoders output 10 and 5 deformation components respectively. For all results, there are some slight deformation components that is not shown for the brief and reasonable visualization.…”
Section: Qualitative Evaluationmentioning
confidence: 99%
“…Since methods compared in the paper produce 50 deformation components, we let K z0 × K z1 = 50. We have the following combinations: K z = [1,50], [2,25], [5,10], [10,5], [25,2], [50,1]. However, the [1,50], [50,1] are the trivial setting.…”
Section: The Choice Of K Zmentioning
confidence: 99%
“…Multi-scale techniques are getting increasingly popular in various fields. In Finite Element Methods, multiscale analysis is widely used [11], [12], [13]. In the spectral geometry field, research works apply multiscale technology on the deformation representation [14], physics-based simulation of deformable objects [15], and surface registration [16] by analyzing the nonisometric global and local (multiscale) deformation.…”
Deformation component analysis is a fundamental problem in geometry processing and shape understanding. Existing approaches mainly extract deformation components in local regions at a similar scale while deformations of real-world objects are usually distributed in a multi-scale manner. In this paper, we propose a novel method to exact multiscale deformation components automatically with a stacked attention-based autoencoder. The attention mechanism is designed to learn to softly weight multi-scale deformation components in active deformation regions, and the stacked attention-based autoencoder is learned to represent the deformation components at different scales. Quantitative and qualitative evaluations show that our method outperforms state-of-the-art methods. Furthermore, with the multiscale deformation components extracted by our method, the user can edit shapes in a coarse-to-fine fashion which facilitates effective modeling of new shapes.
This paper presents a novel approach to concurrent multiscale analysis, where structures are formulated at both microscopic and macro levels for simulation purposes. The proposed method employs a plate model to formulate structures at both scales, and homogenisation is performed using the FFT-based approach, offering higher efficiency compared to conventional methods. Additionally, the macroscopic tangent operator of the microscopic model is derived through an algorithmically consistent process within the FFT-based framework, incorporating the application of Lippman–Schwinger equations as outlined in this work. The effectiveness of the proposed method is demonstrated through case studies in real simulations, revealing comparable results to traditional multiscale schemes in addressing multiscale thin plate structures. Importantly, the method significantly reduces computing time and memory usage, attributed to the efficiency of plate modelling and the FFT-based homogenisation strategy.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.