Pub Date : 2025-04-24DOI: 10.1016/j.gmod.2025.101264
Pengbo Bo , Siyu Xue , Xiwen Xu , Caiming Zhang
Tool shape selection and path planning are critical for 5-axis CNC flank milling of freeform surfaces, typically addressed using optimization algorithms where initialization plays a pivotal role. Existing approaches rely on user-specified initialization of either tool shapes or motion paths, often resulting in suboptimal outcomes. This paper introduces a fully automated method that simultaneously initializes both tool shapes and motion paths, achieving high-precision machining with efficient surface coverage. Our approach explores a solution space of potential tool axes represented by line segments near the design surface. To efficiently manage the vast number of lines, we integrate space voxelization with a discrete distance field for effective line sampling. A graph-based algorithm generates feasible line sequences for motion paths, while path optimization refines a single tool shape across multiple paths simultaneously. The method identifies optimal tool shapes of various sizes, each paired with corresponding motion paths for multi-pass machining. Experiments on industrial benchmark models and freeform surfaces validate the effectiveness and practicality of the proposed approach.
{"title":"Initialization of cutting tools and milling paths for 5-axis CNC flank milling of freeform surfaces","authors":"Pengbo Bo , Siyu Xue , Xiwen Xu , Caiming Zhang","doi":"10.1016/j.gmod.2025.101264","DOIUrl":"10.1016/j.gmod.2025.101264","url":null,"abstract":"<div><div>Tool shape selection and path planning are critical for 5-axis CNC flank milling of freeform surfaces, typically addressed using optimization algorithms where initialization plays a pivotal role. Existing approaches rely on user-specified initialization of either tool shapes or motion paths, often resulting in suboptimal outcomes. This paper introduces a fully automated method that simultaneously initializes both tool shapes and motion paths, achieving high-precision machining with efficient surface coverage. Our approach explores a solution space of potential tool axes represented by line segments near the design surface. To efficiently manage the vast number of lines, we integrate space voxelization with a discrete distance field for effective line sampling. A graph-based algorithm generates feasible line sequences for motion paths, while path optimization refines a single tool shape across multiple paths simultaneously. The method identifies optimal tool shapes of various sizes, each paired with corresponding motion paths for multi-pass machining. Experiments on industrial benchmark models and freeform surfaces validate the effectiveness and practicality of the proposed approach.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"139 ","pages":"Article 101264"},"PeriodicalIF":2.5,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143870721","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-11DOI: 10.1016/j.gmod.2025.101262
Xufei Guo , Xiao Dong , Juan Cao , Zhonggui Chen
The creation of computational agents capable of generating computer-aided design (CAD) models that rival those produced by professional designers is a pressing challenge in the field of computational design. The key obstacle is the need to generate a large number of realistic and diverse models while maintaining control over the output to a certain degree. Therefore, we propose a novel CAD model generation network called CADTrans which is based on a code tree-guided transformer framework to autoregressively generate CAD construction sequences. Firstly, three regularized discrete codebooks are extracted through vector quantized adversarial learning, with each codebook respectively representing the features of Loop, Profile, and Solid. Secondly, these codebooks are used to normalize a CAD construction sequence into a structured code tree representation which is then used to train a standard transformer network to reconstruct the code tree. Finally, the code tree is used as global information to guide the sketch-and-extrude method to recover the corresponding geometric information, thereby reconstructing the complete CAD model. Extensive experiments demonstrate that CADTrans achieves state-of-the-art performance, generating higher-quality, more varied, and complex models. Meanwhile, it provides more possibilities for CAD applications through its flexible control method, enabling users to quickly experiment with different design schemes, inspiring diverse design ideas and the generation of a wide variety of models or even inspiring models, thereby improving design efficiency and promoting creativity. The code is available at https://effieguoxufei.github.io/CADtrans/.
{"title":"CADTrans: A code tree-guided CAD generative transformer model with regularized discrete codebooks","authors":"Xufei Guo , Xiao Dong , Juan Cao , Zhonggui Chen","doi":"10.1016/j.gmod.2025.101262","DOIUrl":"10.1016/j.gmod.2025.101262","url":null,"abstract":"<div><div>The creation of computational agents capable of generating computer-aided design (CAD) models that rival those produced by professional designers is a pressing challenge in the field of computational design. The key obstacle is the need to generate a large number of realistic and diverse models while maintaining control over the output to a certain degree. Therefore, we propose a novel CAD model generation network called CADTrans which is based on a code tree-guided transformer framework to autoregressively generate CAD construction sequences. Firstly, three regularized discrete codebooks are extracted through vector quantized adversarial learning, with each codebook respectively representing the features of Loop, Profile, and Solid. Secondly, these codebooks are used to normalize a CAD construction sequence into a structured code tree representation which is then used to train a standard transformer network to reconstruct the code tree. Finally, the code tree is used as global information to guide the sketch-and-extrude method to recover the corresponding geometric information, thereby reconstructing the complete CAD model. Extensive experiments demonstrate that CADTrans achieves state-of-the-art performance, generating higher-quality, more varied, and complex models. Meanwhile, it provides more possibilities for CAD applications through its flexible control method, enabling users to quickly experiment with different design schemes, inspiring diverse design ideas and the generation of a wide variety of models or even inspiring models, thereby improving design efficiency and promoting creativity. The code is available at <span><span>https://effieguoxufei.github.io/CADtrans/</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"139 ","pages":"Article 101262"},"PeriodicalIF":2.5,"publicationDate":"2025-04-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817456","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-10DOI: 10.1016/j.gmod.2025.101263
Tan Gui , Zhihong Li , Yongjun Cao , Jianghong Yang , Yingjun Wang
This study proposes an efficient preprocessing method and parametric modeling technique for the path planning of corrugated curved surface sandwich structures. Focusing on the characteristics of Fused Deposition Modeling (FDM), the model undergoes preprocessing for two types of path planning, segmenting the sandwich structure for Eulerian Path Printing (EPP) and Eulerian Circuit Printing (ECP). Algorithms were developed using the SolidWorks API for secondary development, resulting in a standalone plugin module. This plugin streamlines adaptive modeling of corrugated sandwich structures on curved surfaces, showcasing strong versatility. Additionally, a comparison of the printing time between preprocessed models and standard models reveals a significant reduction in nozzle idle time. Moreover, as the infill density increases, the reduction in printing time becomes more pronounced. Finally, compression tests confirmed that printed parts obtained using the EPP and ECP methods maintained comparable mechanical properties to those printed using conventional methods.
{"title":"An efficient parametric modeling and path planning method for 3D printing of curved surface corrugated sandwich structures","authors":"Tan Gui , Zhihong Li , Yongjun Cao , Jianghong Yang , Yingjun Wang","doi":"10.1016/j.gmod.2025.101263","DOIUrl":"10.1016/j.gmod.2025.101263","url":null,"abstract":"<div><div>This study proposes an efficient preprocessing method and parametric modeling technique for the path planning of corrugated curved surface sandwich structures. Focusing on the characteristics of Fused Deposition Modeling (FDM), the model undergoes preprocessing for two types of path planning, segmenting the sandwich structure for Eulerian Path Printing (EPP) and Eulerian Circuit Printing (ECP). Algorithms were developed using the SolidWorks API for secondary development, resulting in a standalone plugin module. This plugin streamlines adaptive modeling of corrugated sandwich structures on curved surfaces, showcasing strong versatility. Additionally, a comparison of the printing time between preprocessed models and standard models reveals a significant reduction in nozzle idle time. Moreover, as the infill density increases, the reduction in printing time becomes more pronounced. Finally, compression tests confirmed that printed parts obtained using the EPP and ECP methods maintained comparable mechanical properties to those printed using conventional methods.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"139 ","pages":"Article 101263"},"PeriodicalIF":2.5,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808721","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-04-02DOI: 10.1016/j.gmod.2025.101261
Ao Zhang, Qing Fang, Peng Zhou, Xiao-Ming Fu
Computing the Laplace–Beltrami operator on point clouds is essential for tasks such as smoothing and shape analysis. Unlike meshes, determining the Laplace–Beltrami operator on point clouds requires establishing neighbors for each point. However, traditional -nearest neighbors (k-NN) methods for estimating local neighborhoods often introduce spurious connectivities that distort the manifold topology. We propose a novel approach that leverages persistent homology to refine the neighborhood graph by identifying and removing erroneous edges. Starting with an initial k-NN graph, we assign weights based on local tangent plane estimations and construct a Vietoris–Rips complex. Persistent homology is then employed to detect and eliminate spurious edges through a topological optimization process. This iterative refinement results in a more accurate neighborhood graph that better represents the underlying manifold, enabling precise discretization of the Laplace–Beltrami operator. Experimental results on various point cloud datasets demonstrate that our method outperforms traditional k-NN approaches by more accurately capturing the manifold topology and enhancing downstream computations such as spectral analysis.
{"title":"Topology-controlled Laplace–Beltrami operator on point clouds based on persistent homology","authors":"Ao Zhang, Qing Fang, Peng Zhou, Xiao-Ming Fu","doi":"10.1016/j.gmod.2025.101261","DOIUrl":"10.1016/j.gmod.2025.101261","url":null,"abstract":"<div><div>Computing the Laplace–Beltrami operator on point clouds is essential for tasks such as smoothing and shape analysis. Unlike meshes, determining the Laplace–Beltrami operator on point clouds requires establishing neighbors for each point. However, traditional <span><math><mi>k</mi></math></span>-nearest neighbors (k-NN) methods for estimating local neighborhoods often introduce spurious connectivities that distort the manifold topology. We propose a novel approach that leverages persistent homology to refine the neighborhood graph by identifying and removing erroneous edges. Starting with an initial k-NN graph, we assign weights based on local tangent plane estimations and construct a Vietoris–Rips complex. Persistent homology is then employed to detect and eliminate spurious edges through a topological optimization process. This iterative refinement results in a more accurate neighborhood graph that better represents the underlying manifold, enabling precise discretization of the Laplace–Beltrami operator. Experimental results on various point cloud datasets demonstrate that our method outperforms traditional k-NN approaches by more accurately capturing the manifold topology and enhancing downstream computations such as spectral analysis.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"139 ","pages":"Article 101261"},"PeriodicalIF":2.5,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143746699","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-28DOI: 10.1016/j.gmod.2025.101260
Gershon Elber
Compliant mechanisms have captured the attention of many researchers in recent years, with ever-expanding mechanical and physical behaviors. Moreover, the fabrication of such mechanisms has been greatly simplified with the enabling technology of 3D printing.
Drawing from existing lattice construction abilities, in this work we explore the abilities of constructing whole freeform lattices, where the tiles in these lattices are compliant mechanisms, possibly heterogeneous. Specifically, herein we focus on bi-stable and multi-stable tiles, or tiles with two or more mechanical stable states. The introduced approach will be exemplified on a variety of 2D and 3D lattices, fabricated with the aid of additive manufacturing.
{"title":"Additive manufacturing toward 2D and 3D freeform lattices with conforming compliant bi-/multi-stable tiles","authors":"Gershon Elber","doi":"10.1016/j.gmod.2025.101260","DOIUrl":"10.1016/j.gmod.2025.101260","url":null,"abstract":"<div><div>Compliant mechanisms have captured the attention of many researchers in recent years, with ever-expanding mechanical and physical behaviors. Moreover, the fabrication of such mechanisms has been greatly simplified with the enabling technology of 3D printing.</div><div>Drawing from existing lattice construction abilities, in this work we explore the abilities of constructing whole freeform lattices, where the tiles in these lattices are compliant mechanisms, possibly heterogeneous. Specifically, herein we focus on bi-stable and multi-stable tiles, or tiles with two or more mechanical stable states. The introduced approach will be exemplified on a variety of 2D and 3D lattices, fabricated with the aid of additive manufacturing.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"139 ","pages":"Article 101260"},"PeriodicalIF":2.5,"publicationDate":"2025-03-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143716109","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-25DOI: 10.1016/j.gmod.2025.101257
Jiaming Peng, Xinhai Chen, Jie Liu
Mesh generation is a crucial step in numerical simulations, significantly impacting simulation accuracy and efficiency. However, generating meshes remains time-consuming and requires expensive computational resources. In this paper, we propose a novel method, 3DMeshNet, for three-dimensional structured mesh generation. The method embeds the meshing-related differential equations into the loss function of neural networks, formulating the meshing task as an unsupervised optimization problem. It takes geometric points as input to learn the potential mapping between parametric and computational spaces. After suitable offline training, 3DMeshNet can efficiently output a three-dimensional structured mesh with a user-defined number of quadrilateral/hexahedral cells through the feed-forward neural prediction. To enhance training stability and accelerate convergence, we integrate loss function reweighting through weight adjustments and gradient projection alongside applying finite difference methods to streamline derivative computations in the loss. Experiments on different cases show that 3DMeshNet is robust and fast. It outperforms neural network-based methods and yields superior meshes compared to traditional mesh partitioning methods. 3DMeshNet significantly reduces training times by up to 85% compared to other neural network-based approaches and lowers meshing overhead by 4 to 8 times relative to traditional meshing methods.
{"title":"3DMeshNet: A three-dimensional differential neural network for structured mesh generation","authors":"Jiaming Peng, Xinhai Chen, Jie Liu","doi":"10.1016/j.gmod.2025.101257","DOIUrl":"10.1016/j.gmod.2025.101257","url":null,"abstract":"<div><div>Mesh generation is a crucial step in numerical simulations, significantly impacting simulation accuracy and efficiency. However, generating meshes remains time-consuming and requires expensive computational resources. In this paper, we propose a novel method, 3DMeshNet, for three-dimensional structured mesh generation. The method embeds the meshing-related differential equations into the loss function of neural networks, formulating the meshing task as an unsupervised optimization problem. It takes geometric points as input to learn the potential mapping between parametric and computational spaces. After suitable offline training, 3DMeshNet can efficiently output a three-dimensional structured mesh with a user-defined number of quadrilateral/hexahedral cells through the feed-forward neural prediction. To enhance training stability and accelerate convergence, we integrate loss function reweighting through weight adjustments and gradient projection alongside applying finite difference methods to streamline derivative computations in the loss. Experiments on different cases show that 3DMeshNet is robust and fast. It outperforms neural network-based methods and yields superior meshes compared to traditional mesh partitioning methods. 3DMeshNet significantly reduces training times by up to 85% compared to other neural network-based approaches and lowers meshing overhead by 4 to 8 times relative to traditional meshing methods.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"139 ","pages":"Article 101257"},"PeriodicalIF":2.5,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143683472","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-25DOI: 10.1016/j.gmod.2025.101259
Wenjin Yang, Jie He, Xiaotong Zhang
Line charts, as a common data visualization tool in scientific research and business analysis, encapsulate rich experimental data. However, existing data extraction tools face challenges such as low automation levels and difficulties in handling complex charts. This paper proposes a novel method for extracting data from line charts, reformulating the extraction problem as an instance segmentation task, and introducing the Mamba-enhanced Transformer mask query method along with a curve mask-guided training approach to address challenges such as long dependencies and intersections in curve detection. Additionally, YOLOv9 is utilized for the detection and classification of chart elements, and a text recognition dataset comprising approximately 100K charts is constructed. An LSTM-based attention mechanism is employed for precise scale value recognition. Lastly, we present a method for automatically converting image data into structured JSON data, significantly enhancing the efficiency and accuracy of data extraction. Experimental results demonstrate that this method exhibits high efficiency and accuracy in handling complex charts, achieving an average extraction accuracy of 93% on public datasets, significantly surpassing the current state-of-the-art methods. This research provides an efficient foundation for large-scale scientific data analysis and machine learning model development, advancing the field of automated data extraction technology.
{"title":"Efficient extraction of experimental data from line charts using advanced machine learning techniques","authors":"Wenjin Yang, Jie He, Xiaotong Zhang","doi":"10.1016/j.gmod.2025.101259","DOIUrl":"10.1016/j.gmod.2025.101259","url":null,"abstract":"<div><div>Line charts, as a common data visualization tool in scientific research and business analysis, encapsulate rich experimental data. However, existing data extraction tools face challenges such as low automation levels and difficulties in handling complex charts. This paper proposes a novel method for extracting data from line charts, reformulating the extraction problem as an instance segmentation task, and introducing the Mamba-enhanced Transformer mask query method along with a curve mask-guided training approach to address challenges such as long dependencies and intersections in curve detection. Additionally, YOLOv9 is utilized for the detection and classification of chart elements, and a text recognition dataset comprising approximately 100K charts is constructed. An LSTM-based attention mechanism is employed for precise scale value recognition. Lastly, we present a method for automatically converting image data into structured JSON data, significantly enhancing the efficiency and accuracy of data extraction. Experimental results demonstrate that this method exhibits high efficiency and accuracy in handling complex charts, achieving an average extraction accuracy of 93% on public datasets, significantly surpassing the current state-of-the-art methods. This research provides an efficient foundation for large-scale scientific data analysis and machine learning model development, advancing the field of automated data extraction technology.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"139 ","pages":"Article 101259"},"PeriodicalIF":2.5,"publicationDate":"2025-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143683471","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-24DOI: 10.1016/j.gmod.2025.101258
Hongyu Chen, Xiaodiao Chen, Yizhao Xue
Medial axis computation has wide applications in pattern recognition, image processing, finite element mesh generation, and CNC tool path extraction. Aiming to explore intrinsic geometric attributes of the medial axis of a simple polygon which can be accurately represented and faces its challenge of computational efficiency, an R-L sequence-based algorithm of linear computational complexity is proposed for achieving much higher efficiency; especially, it enables the complexity of Delaunay triangulation to be linear. The algorithm is done by reconstructing the Voronoi diagram tree of the given simple polygon, which can be easily performed in a breadth-first manner with a higher computational efficiency. The branches of the medial axis are naturally divided into several panels, such that the branches in the same panel cause no interference with each other and decrease a lot of computational costs. Based on our experiments, the efficiency of the proposed R-L algorithm can be 6 to 17 times greater than that of the state-of-the-art method in TVCG, and up to 419 times greater than the CGAL algorithm. In principle, it can be directly applied to compute the medial axis of curvilinear polygons, which expands the scope of application compared to Chin’s method.
{"title":"Computing medial axis of a simple polygon in linear time based on R-L sequence","authors":"Hongyu Chen, Xiaodiao Chen, Yizhao Xue","doi":"10.1016/j.gmod.2025.101258","DOIUrl":"10.1016/j.gmod.2025.101258","url":null,"abstract":"<div><div>Medial axis computation has wide applications in pattern recognition, image processing, finite element mesh generation, and CNC tool path extraction. Aiming to explore intrinsic geometric attributes of the medial axis of a simple polygon which can be accurately represented and faces its challenge of computational efficiency, an R-L sequence-based algorithm of linear computational complexity is proposed for achieving much higher efficiency; especially, it enables the complexity of Delaunay triangulation to be linear. The algorithm is done by reconstructing the Voronoi diagram tree of the given simple polygon, which can be easily performed in a breadth-first manner with a higher computational efficiency. The branches of the medial axis are naturally divided into several panels, such that the branches in the same panel cause no interference with each other and decrease a lot of computational costs. Based on our experiments, the efficiency of the proposed R-L algorithm can be 6 to 17 times greater than that of the state-of-the-art method in TVCG, and up to 419 times greater than the CGAL algorithm. In principle, it can be directly applied to compute the medial axis of curvilinear polygons, which expands the scope of application compared to Chin’s method.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"139 ","pages":"Article 101258"},"PeriodicalIF":2.5,"publicationDate":"2025-03-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143683549","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-04DOI: 10.1016/j.gmod.2025.101255
Changshuang Zhou , Frederick W.B. Li , Chao Song , Dong Zheng , Bailin Yang
We propose Dual-Branch Network (DBNet), a novel deepfake detection framework that addresses key limitations of existing works by jointly modeling 3D-temporal and fine-grained texture representations. Specifically, we aim to investigate how to (1) capture dynamic properties and spatial details in a unified model and (2) identify subtle inconsistencies beyond localized artifacts through temporally consistent modeling. To this end, DBNet extracts 3D landmarks from videos to construct temporal sequences for an RNN branch, while a Vision Transformer analyzes local patches. A Temporal Consistency-aware Loss is introduced to explicitly supervise the RNN. Additionally, a 3D generative model augments training data. Extensive experiments demonstrate our method achieves state-of-the-art performance on benchmarks, and ablation studies validate its effectiveness in generalizing to unseen data under various manipulations and compression.
{"title":"3D data augmentation and dual-branch model for robust face forgery detection","authors":"Changshuang Zhou , Frederick W.B. Li , Chao Song , Dong Zheng , Bailin Yang","doi":"10.1016/j.gmod.2025.101255","DOIUrl":"10.1016/j.gmod.2025.101255","url":null,"abstract":"<div><div>We propose Dual-Branch Network (DBNet), a novel deepfake detection framework that addresses key limitations of existing works by jointly modeling 3D-temporal and fine-grained texture representations. Specifically, we aim to investigate how to (1) capture dynamic properties and spatial details in a unified model and (2) identify subtle inconsistencies beyond localized artifacts through temporally consistent modeling. To this end, DBNet extracts 3D landmarks from videos to construct temporal sequences for an RNN branch, while a Vision Transformer analyzes local patches. A Temporal Consistency-aware Loss is introduced to explicitly supervise the RNN. Additionally, a 3D generative model augments training data. Extensive experiments demonstrate our method achieves state-of-the-art performance on benchmarks, and ablation studies validate its effectiveness in generalizing to unseen data under various manipulations and compression.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"138 ","pages":"Article 101255"},"PeriodicalIF":2.5,"publicationDate":"2025-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143104630","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-02-01DOI: 10.1016/j.gmod.2024.101253
Manuel Prado-Velasco, Laura García-Ruesgas
The Computer extended Descriptive Geometry (CeDG) is as a novel approach based on Descriptive Geometry to build 3D models within the framework provided by Dynamic Geometry Software tools. Parametric CeDG models can be interactively explored when continuous parameters change, but this is not the case for discrete parameters. This study demonstrates the capability of the GeoGebra - CeDG approach to incorporate algorithms that build discrete variable 3D models with dynamic parameterization. Several 3D models and their flattened patterns (neutral fiber), based on a new developed CeDG algorithm, were compared to their LogiTRACE v.14 and Solid Edge 2024 (CAD) counterparts. The accuracy of the CeDG models surpassed that of CAD models for nearly all dimensions defined as metrics. In addition, the CeDG approach was the unique that provided an automatic solution for any value of the number of ferrules.
{"title":"Discrete variable 3D models in Computer extended Descriptive Geometry (CeDG): Building of polygonal sheet-metal elbows and comparison against CAD","authors":"Manuel Prado-Velasco, Laura García-Ruesgas","doi":"10.1016/j.gmod.2024.101253","DOIUrl":"10.1016/j.gmod.2024.101253","url":null,"abstract":"<div><div>The Computer extended Descriptive Geometry (CeDG) is as a novel approach based on Descriptive Geometry to build 3D models within the framework provided by Dynamic Geometry Software tools. Parametric CeDG models can be interactively explored when continuous parameters change, but this is not the case for discrete parameters. This study demonstrates the capability of the GeoGebra - CeDG approach to incorporate algorithms that build discrete variable 3D models with dynamic parameterization. Several 3D models and their flattened patterns (neutral fiber), based on a new developed CeDG algorithm, were compared to their LogiTRACE v.14 and Solid Edge 2024 (CAD) counterparts. The accuracy of the CeDG models surpassed that of CAD models for nearly all dimensions defined as metrics. In addition, the CeDG approach was the unique that provided an automatic solution for any value of the number of ferrules.</div></div>","PeriodicalId":55083,"journal":{"name":"Graphical Models","volume":"137 ","pages":"Article 101253"},"PeriodicalIF":2.5,"publicationDate":"2025-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143140781","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号