Objectives： The sails fabrication knew a significant development under the impulse of sailing races like the America＇s Cup and the Volvo around the World Race. These competitions require advanced technologies to inc...Objectives： The sails fabrication knew a significant development under the impulse of sailing races like the America＇s Cup and the Volvo around the World Race. These competitions require advanced technologies to increase performance of the sails. In the paper, Reverse Engineering （RE） acquisition, CAD （Computer Aided Design） reconstruction and FE （Finite Element） analysis are used in the design phase for a spinnaker in order to maximize the performance of the sail boat. Materials and Methods： The comparison, between the design data and the characteristics in the real conditions of use of the sail （constant wind pressure acting on the sail）, was carried out by the reconstruction of the three-dimensional CAD model of the spinnaker shape in full-scale, using a non contact Reverse Engineering technique, the photogrammetry and by a FE non linear static analysis. Results： The results of this study show a good correspondence between the design data and the characteristics in the real conditions of use of the sail. Significance： The described methodology is useful for sail makers. It can be successfully adopted to generate three-dimensional CAD-FE models of a sail starting from a very cheap and fast Reverse Engineering non-contact procedure.展开更多
Existing reverse-engineering methods struggle to directly generate editable,parametric CAD models from scanned data.To address this limitation,this paper proposes a reverse-modeling approach that reconstructs parametr...Existing reverse-engineering methods struggle to directly generate editable,parametric CAD models from scanned data.To address this limitation,this paper proposes a reverse-modeling approach that reconstructs parametric CAD models from multi-view RGB-D point clouds.Multi-frame point-cloud registration and fusion are first employed to obtain a complete 3-D point cloud of the target object.A region-growing algorithm that jointly exploits color and geometric information segments the cloud,while RANSAC robustly detects and fits basic geometric primitives.These primitives serve as nodes in a graph whose edge features are inferred by a graph neural network to capture spatial constraints.From the detected primitives and their constraints,a high-accuracy,fully editable parametric CAD model is finally exported.Experiments show an average parameter error of 0.3 mm for key dimensions and an overall geometric reconstruction accuracy of 0.35 mm.The work offers an effective technical route toward automated,intelligent 3-D reverse modeling.展开更多
The current computer-aided technologies in design and product development, the evolution of CAD modeling, and a framework of multi-volume CADmodeling system for heterogeneous object design and fabrication are presente...The current computer-aided technologies in design and product development, the evolution of CAD modeling, and a framework of multi-volume CADmodeling system for heterogeneous object design and fabrication are presented inthis paper. The multi-volume CAD modeling system is presented based on nonmanifold topological elements. Material identifications are defined as design attributes introduced along with geometric and topological information at the designstage. Extended Euler operation and reasoning Boolean operations for merging andextraction are executed according to the associated material identifications in thedeveloped multi-volume modeling system for heterogeneous object. An applicationexample and a pseudo-processing algorithm for prototyping of heterogeneous structure through solid free-form fabrication are also described.展开更多
In this paper,we introduce an innovative method for computer-aided design(CAD)segmentation by concatenating meshes and CAD models.Many previous CAD segmentation methods have achieved impressive performance using singl...In this paper,we introduce an innovative method for computer-aided design(CAD)segmentation by concatenating meshes and CAD models.Many previous CAD segmentation methods have achieved impressive performance using single representations,such as meshes,CAD,and point clouds.However,existing methods cannot effectively combine different three-dimensional model types for the direct conversion,alignment,and integrity maintenance of geometric and topological information.Hence,we propose an integration approach that combines the geometric accuracy of CAD data with the flexibility of mesh representations,as well as introduce a unique hybrid representation that combines CAD and mesh models to enhance segmentation accuracy.To combine these two model types,our hybrid system utilizes advanced-neural-network techniques to convert CAD models into mesh models.For complex CAD models,model segmentation is crucial for model retrieval and reuse.In partial retrieval,it aims to segment a complex CAD model into several simple components.The first component of our hybrid system involves advanced mesh-labeling algorithms that harness the digitization of CAD properties to mesh models.The second component integrates labelled face features for CAD segmentation by leveraging the abundant multisemantic information embedded in CAD models.This combination of mesh and CAD not only refines the accuracy of boundary delineation but also provides a comprehensive understanding of the underlying object semantics.This study uses the Fusion 360 Gallery dataset.Experimental results indicate that our hybrid method can segment these models with higher accuracy than other methods that use single representations.展开更多
Cloud manufacturing is one of the three key technologies that enable intelligent manufacturing.This paper presents a novel attribute-based encryption(ABE)approach for computer-aided design(CAD)assembly models to effec...Cloud manufacturing is one of the three key technologies that enable intelligent manufacturing.This paper presents a novel attribute-based encryption(ABE)approach for computer-aided design(CAD)assembly models to effectively support hierarchical access control,integrity verification,and deformation protection for co-design scenarios in cloud manufacturing.An assembly hierarchy access tree(AHAT)is designed as the hierarchical access structure.Attribute-related ciphertext elements,which are contained in an assembly ciphertext(ACT)file,are adapted for content keys decryption instead of CAD component files.We modify the original Merkle tree(MT)and reconstruct an assembly MT.The proposed ABE framework has the ability to combine the deformation protection method with a content privacy of CAD models.The proposed encryption scheme is demonstrated to be secure under the standard assumption.Experimental simulation on typical CAD assembly models demonstrates that the proposed approach is feasible in applications.展开更多
CAD model retrieval based on functional semantics is more significant than content-based 3D model retrieval during the mechanical conceptual design phase. However, relevant research is still not fully discussed. There...CAD model retrieval based on functional semantics is more significant than content-based 3D model retrieval during the mechanical conceptual design phase. However, relevant research is still not fully discussed. Therefore, a functional semantic-based CAD model annotation and retrieval method is proposed to support mechanical conceptual design and design reuse, inspire designer creativity through existing CAD models, shorten design cycle, and reduce costs. Firstly, the CAD model functional semantic ontology is constructed to formally represent the functional semantics of CAD models and describe the mechanical conceptual design space comprehensively and consistently. Secondly, an approach to represent CAD models as attributed adjacency graphs（AAG） is proposed. In this method, the geometry and topology data are extracted from STEP models. On the basis of AAG, the functional semantics of CAD models are annotated semi-automatically by matching CAD models that contain the partial features of which functional semantics have been annotated manually, thereby constructing CAD Model Repository that supports model retrieval based on functional semantics. Thirdly, a CAD model retrieval algorithm that supports multi-function extended retrieval is proposed to explore more potential creative design knowledge in the semantic level. Finally, a prototype system, called Functional Semantic-based CAD Model Annotation and Retrieval System（FSMARS）, is implemented. A case demonstrates that FSMARS can successfully botain multiple potential CAD models that conform to the desired function. The proposed research addresses actual needs and presents a new way to acquire CAD models in the mechanical conceptual design phase.展开更多
Based on the principles of differential geometry and kinematics, a mathematical model is developed to describe the grinding wheel axial cross-section with the radial cross-section of the flute in a given drill under t...Based on the principles of differential geometry and kinematics, a mathematical model is developed to describe the grinding wheel axial cross-section with the radial cross-section of the flute in a given drill under the basic engagement condition between the generating flute and the generated grinding wheel （or disk milling tool）. The mathematical model is good for the flute in the radial cross-section consisting of three arcs. Furthermore, a CAD system is also developed to represent the axial cross-section of the grinding wheel （or disk milling tool）. With the system, the grinding wheel （or disk milling tool） axial cross-section that corresponds to the three-arc flute cross section can be conveniently simulated. Through the grinding experiment of drill flutes, the method and the CAD system are proved to be feasible and reasonable.展开更多
We present the solid model edit distance(SMED),a powerful and flexible paradigm for exploiting shape similarities amongst CAD models.It is designed to measure the magnitude of distortions between two CAD models in bou...We present the solid model edit distance(SMED),a powerful and flexible paradigm for exploiting shape similarities amongst CAD models.It is designed to measure the magnitude of distortions between two CAD models in boundary representation(B-rep).We give the formal definition by analogy with graph edit distance,one of the most popular graph matching methods.To avoid the expensive computational cost potentially caused by exact computation,an approximate procedure based on the alignment of local structure sets is provided in addition.In order to verify the flexibility,we make intensive investigations on three typical applications in manufacturing industry,and describe how our method can be adapted to meet the various requirements.Furthermore,a multilevel method is proposed to make further improvements of the presented algorithm on both effectiveness and efficiency,in which the models are hierarchically segmented into the configurations of features.Experiment results show that SMED serves as a reasonable measurement of shape similarity for CAD models,and the proposed approach provides remarkable performance on a real-world CAD model database.展开更多
An algorithm for compensating the varying section contour within the extent of each layer thickness is proposed to improve the rapid prototyping part’s shape and size precision.
To improve the reusability of three-dimensional （3D） models and simplify the complexity of natural scene reconstruction, this paper presents a 3D model database for universal 3D GIS. After the introduction of its ex...To improve the reusability of three-dimensional （3D） models and simplify the complexity of natural scene reconstruction, this paper presents a 3D model database for universal 3D GIS. After the introduction of its extensible function architecture, accompanied by the conclusion of implicit spatial-temporal hierarchy of models in any reconstructed scene of 3D GIS for general purpose, several key issues are discussed in detail, such as the storage and management of 3D models and related retrieval and load method, as well as the interfaces for further on-demand development. Finally, the validity and feasibility of this model database are proved through its application in the development of 3D visualization system of railway operation.展开更多
The geometrical and topological information of 3D computer aided design （CAD） models should be represented as a neut- ral format file to exchange the data between different CAD systems. Exchange of 3D CAD model data...The geometrical and topological information of 3D computer aided design （CAD） models should be represented as a neut- ral format file to exchange the data between different CAD systems. Exchange of 3D CAD model data implies that the companies must exchange complete information about their products, all the way from design, manufacturing to inspection and shipping. This informa- tion should be available to each relevant partner over the entire life cycle of the product. This led to the development of an international standard organization （ISO） neutral format file named as standard for the exchange of product model data （STEP）. It has been ob- served from the literature, the feature recognition systems developed were identified as planar, cylindrical, conical and to some extent spherical and toroidal surfaces. The advanced surface features such as B-spline and its subtypes are not identified. Therefore, in this work, a STEP-based feature recognition system is developed to recognize t--spline surface features and its sub-types from the 3D CAD model represented in AP203 neutral file format. The developed feature recognition system is implemented in Java programming language and the product model data represented in STEP AP203 format is interpreted through Java standard data access interface （JSDAI）. The developed system could recognize B-spline surface features such as B-Spline surface with knots, quasi uniform surface, uniform surface, rational surface and Bezier surface. The application of extracted B-spline surface features information is discussed with reference to the toolpath generation for STEP-NC （STEP AP238）.展开更多
Three-dimensional(3D)tracking of rigid objects plays a very important role in many areas such as augmented reality,computer vision,and robotics.Numerous works have been done to pursue more stable,faster,and more accur...Three-dimensional(3D)tracking of rigid objects plays a very important role in many areas such as augmented reality,computer vision,and robotics.Numerous works have been done to pursue more stable,faster,and more accurate 3D tracking.Among various tracking methods,edge-based 3D tracking has been widely used owing to its many advantages.Furthermore,edge-based methods can be mainly divided into two categories,methods without and those with explicit edges,depending on whether explicit edges need to be extracted.Based on this,representative methods in both categories are introduced,analyzed,and compared in this paper.Finally,some suggestions on the choice of methods in different application scenarios and research directions in the future are given.展开更多
Reconstructing from multi-view images is a longstanding problem in 3D vision,where neural radiance fields(NeRFs)have shown great potential and get realistic rendered images of novel views.Currently,most NeRF methods e...Reconstructing from multi-view images is a longstanding problem in 3D vision,where neural radiance fields(NeRFs)have shown great potential and get realistic rendered images of novel views.Currently,most NeRF methods either require accurate camera poses or a large number of input images,or even both.Reconstructing NeRF from few-view images without poses is challenging and highly ill-posed.To address this problem,we propose CAD-NeRF,a method reconstructed from less than 10 images without any known poses.Specifically,we build a mini library of several CAD models from ShapeNet and render them from many random views.Given sparse-view input images,we run a model and pose retrieval from the library,to get a model with similar shapes,serving as the density supervision and pose initializations.Here we propose a multi-view pose retrieval method to avoid pose conflicts among views,which is a new and unseen problem in uncalibrated NeRF methods.Then,the geometry of the object is trained by the CAD guidance.The deformation of the density field and camera poses are optimized jointly.Then texture and density are trained and fine-tuned as well.All training phases are in self-supervised manners.Comprehensive evaluations of synthetic and real images show that CAD-NeRF successfully learns accurate densities with a large deformation from retrieved CAD models,showing the generalization abilities.展开更多
Model classification is essential to the management and reuse of 3D CAD models.Manual model classification is laborious and error prone.At the same time,the automatic classification methods are scarce due to the intri...Model classification is essential to the management and reuse of 3D CAD models.Manual model classification is laborious and error prone.At the same time,the automatic classification methods are scarce due to the intrinsic complexity of 3D CAD models.In this paper,we propose an automatic 3D CAD model classification approach based on deep neural networks.According to prior knowledge of the CAD domain,features are selected and extracted from 3D CAD models first,and then preprocessed as high dimensional input vectors for category recognition.By analogy with the thinking process of engineers,a deep neural network classifier for 3D CAD models is constructed with the aid of deep learning techniques.To obtain an optimal solution,multiple strategies are appropriately chosen and applied in the training phase,which makes our classifier achieve better performance.We demonstrate the efficiency and effectiveness of our approach through experiments on 3D CAD model datasets.展开更多
文摘Objectives： The sails fabrication knew a significant development under the impulse of sailing races like the America＇s Cup and the Volvo around the World Race. These competitions require advanced technologies to increase performance of the sails. In the paper, Reverse Engineering （RE） acquisition, CAD （Computer Aided Design） reconstruction and FE （Finite Element） analysis are used in the design phase for a spinnaker in order to maximize the performance of the sail boat. Materials and Methods： The comparison, between the design data and the characteristics in the real conditions of use of the sail （constant wind pressure acting on the sail）, was carried out by the reconstruction of the three-dimensional CAD model of the spinnaker shape in full-scale, using a non contact Reverse Engineering technique, the photogrammetry and by a FE non linear static analysis. Results： The results of this study show a good correspondence between the design data and the characteristics in the real conditions of use of the sail. Significance： The described methodology is useful for sail makers. It can be successfully adopted to generate three-dimensional CAD-FE models of a sail starting from a very cheap and fast Reverse Engineering non-contact procedure.
文摘Existing reverse-engineering methods struggle to directly generate editable,parametric CAD models from scanned data.To address this limitation,this paper proposes a reverse-modeling approach that reconstructs parametric CAD models from multi-view RGB-D point clouds.Multi-frame point-cloud registration and fusion are first employed to obtain a complete 3-D point cloud of the target object.A region-growing algorithm that jointly exploits color and geometric information segments the cloud,while RANSAC robustly detects and fits basic geometric primitives.These primitives serve as nodes in a graph whose edge features are inferred by a graph neural network to capture spatial constraints.From the detected primitives and their constraints,a high-accuracy,fully editable parametric CAD model is finally exported.Experiments show an average parameter error of 0.3 mm for key dimensions and an overall geometric reconstruction accuracy of 0.35 mm.The work offers an effective technical route toward automated,intelligent 3-D reverse modeling.
文摘The current computer-aided technologies in design and product development, the evolution of CAD modeling, and a framework of multi-volume CADmodeling system for heterogeneous object design and fabrication are presented inthis paper. The multi-volume CAD modeling system is presented based on nonmanifold topological elements. Material identifications are defined as design attributes introduced along with geometric and topological information at the designstage. Extended Euler operation and reasoning Boolean operations for merging andextraction are executed according to the associated material identifications in thedeveloped multi-volume modeling system for heterogeneous object. An applicationexample and a pseudo-processing algorithm for prototyping of heterogeneous structure through solid free-form fabrication are also described.
基金Supported by the National Key Research and Development Program of China(2024YFB3311703)National Natural Science Foundation of China(61932003)Beijing Science and Technology Plan Project(Z221100006322003).
文摘In this paper,we introduce an innovative method for computer-aided design(CAD)segmentation by concatenating meshes and CAD models.Many previous CAD segmentation methods have achieved impressive performance using single representations,such as meshes,CAD,and point clouds.However,existing methods cannot effectively combine different three-dimensional model types for the direct conversion,alignment,and integrity maintenance of geometric and topological information.Hence,we propose an integration approach that combines the geometric accuracy of CAD data with the flexibility of mesh representations,as well as introduce a unique hybrid representation that combines CAD and mesh models to enhance segmentation accuracy.To combine these two model types,our hybrid system utilizes advanced-neural-network techniques to convert CAD models into mesh models.For complex CAD models,model segmentation is crucial for model retrieval and reuse.In partial retrieval,it aims to segment a complex CAD model into several simple components.The first component of our hybrid system involves advanced mesh-labeling algorithms that harness the digitization of CAD properties to mesh models.The second component integrates labelled face features for CAD segmentation by leveraging the abundant multisemantic information embedded in CAD models.This combination of mesh and CAD not only refines the accuracy of boundary delineation but also provides a comprehensive understanding of the underlying object semantics.This study uses the Fusion 360 Gallery dataset.Experimental results indicate that our hybrid method can segment these models with higher accuracy than other methods that use single representations.
基金supported by the National Natural Science Foundation of China(62072348)the Science and Technology Major Project of Hubei Province(Next-Generation AI Technologies,2019AEA170).
文摘Cloud manufacturing is one of the three key technologies that enable intelligent manufacturing.This paper presents a novel attribute-based encryption(ABE)approach for computer-aided design(CAD)assembly models to effectively support hierarchical access control,integrity verification,and deformation protection for co-design scenarios in cloud manufacturing.An assembly hierarchy access tree(AHAT)is designed as the hierarchical access structure.Attribute-related ciphertext elements,which are contained in an assembly ciphertext(ACT)file,are adapted for content keys decryption instead of CAD component files.We modify the original Merkle tree(MT)and reconstruct an assembly MT.The proposed ABE framework has the ability to combine the deformation protection method with a content privacy of CAD models.The proposed encryption scheme is demonstrated to be secure under the standard assumption.Experimental simulation on typical CAD assembly models demonstrates that the proposed approach is feasible in applications.
基金Supported by National Natural Science Foundation of China (Grant No.51175287)National Science and Technology Major Project of China (Grant No.2011ZX02403)
文摘CAD model retrieval based on functional semantics is more significant than content-based 3D model retrieval during the mechanical conceptual design phase. However, relevant research is still not fully discussed. Therefore, a functional semantic-based CAD model annotation and retrieval method is proposed to support mechanical conceptual design and design reuse, inspire designer creativity through existing CAD models, shorten design cycle, and reduce costs. Firstly, the CAD model functional semantic ontology is constructed to formally represent the functional semantics of CAD models and describe the mechanical conceptual design space comprehensively and consistently. Secondly, an approach to represent CAD models as attributed adjacency graphs（AAG） is proposed. In this method, the geometry and topology data are extracted from STEP models. On the basis of AAG, the functional semantics of CAD models are annotated semi-automatically by matching CAD models that contain the partial features of which functional semantics have been annotated manually, thereby constructing CAD Model Repository that supports model retrieval based on functional semantics. Thirdly, a CAD model retrieval algorithm that supports multi-function extended retrieval is proposed to explore more potential creative design knowledge in the semantic level. Finally, a prototype system, called Functional Semantic-based CAD Model Annotation and Retrieval System（FSMARS）, is implemented. A case demonstrates that FSMARS can successfully botain multiple potential CAD models that conform to the desired function. The proposed research addresses actual needs and presents a new way to acquire CAD models in the mechanical conceptual design phase.
基金This project is supported by National Natural Science Foundation of China (No.50675065).
文摘Based on the principles of differential geometry and kinematics, a mathematical model is developed to describe the grinding wheel axial cross-section with the radial cross-section of the flute in a given drill under the basic engagement condition between the generating flute and the generated grinding wheel （or disk milling tool）. The mathematical model is good for the flute in the radial cross-section consisting of three arcs. Furthermore, a CAD system is also developed to represent the axial cross-section of the grinding wheel （or disk milling tool）. With the system, the grinding wheel （or disk milling tool） axial cross-section that corresponds to the three-arc flute cross section can be conveniently simulated. Through the grinding experiment of drill flutes, the method and the CAD system are proved to be feasible and reasonable.
基金Supported by National Science Foundation of China(61373071)
文摘We present the solid model edit distance(SMED),a powerful and flexible paradigm for exploiting shape similarities amongst CAD models.It is designed to measure the magnitude of distortions between two CAD models in boundary representation(B-rep).We give the formal definition by analogy with graph edit distance,one of the most popular graph matching methods.To avoid the expensive computational cost potentially caused by exact computation,an approximate procedure based on the alignment of local structure sets is provided in addition.In order to verify the flexibility,we make intensive investigations on three typical applications in manufacturing industry,and describe how our method can be adapted to meet the various requirements.Furthermore,a multilevel method is proposed to make further improvements of the presented algorithm on both effectiveness and efficiency,in which the models are hierarchically segmented into the configurations of features.Experiment results show that SMED serves as a reasonable measurement of shape similarity for CAD models,and the proposed approach provides remarkable performance on a real-world CAD model database.
基金Supported by College Science Foundation of Jiangsu(07KJB460113)
文摘An algorithm for compensating the varying section contour within the extent of each layer thickness is proposed to improve the rapid prototyping part’s shape and size precision.
基金Supported by the National Natural Science Foundation of China (No.40871212, No.40671158)the National High Technology Research and Development Program of China (No.2008AA121600)
文摘To improve the reusability of three-dimensional （3D） models and simplify the complexity of natural scene reconstruction, this paper presents a 3D model database for universal 3D GIS. After the introduction of its extensible function architecture, accompanied by the conclusion of implicit spatial-temporal hierarchy of models in any reconstructed scene of 3D GIS for general purpose, several key issues are discussed in detail, such as the storage and management of 3D models and related retrieval and load method, as well as the interfaces for further on-demand development. Finally, the validity and feasibility of this model database are proved through its application in the development of 3D visualization system of railway operation.
文摘The geometrical and topological information of 3D computer aided design （CAD） models should be represented as a neut- ral format file to exchange the data between different CAD systems. Exchange of 3D CAD model data implies that the companies must exchange complete information about their products, all the way from design, manufacturing to inspection and shipping. This informa- tion should be available to each relevant partner over the entire life cycle of the product. This led to the development of an international standard organization （ISO） neutral format file named as standard for the exchange of product model data （STEP）. It has been ob- served from the literature, the feature recognition systems developed were identified as planar, cylindrical, conical and to some extent spherical and toroidal surfaces. The advanced surface features such as B-spline and its subtypes are not identified. Therefore, in this work, a STEP-based feature recognition system is developed to recognize t--spline surface features and its sub-types from the 3D CAD model represented in AP203 neutral file format. The developed feature recognition system is implemented in Java programming language and the product model data represented in STEP AP203 format is interpreted through Java standard data access interface （JSDAI）. The developed system could recognize B-spline surface features such as B-Spline surface with knots, quasi uniform surface, uniform surface, rational surface and Bezier surface. The application of extracted B-spline surface features information is discussed with reference to the toolpath generation for STEP-NC （STEP AP238）.
基金Special Program of the Ministry of Industry and Information Technology of China.
文摘Three-dimensional(3D)tracking of rigid objects plays a very important role in many areas such as augmented reality,computer vision,and robotics.Numerous works have been done to pursue more stable,faster,and more accurate 3D tracking.Among various tracking methods,edge-based 3D tracking has been widely used owing to its many advantages.Furthermore,edge-based methods can be mainly divided into two categories,methods without and those with explicit edges,depending on whether explicit edges need to be extracted.Based on this,representative methods in both categories are introduced,analyzed,and compared in this paper.Finally,some suggestions on the choice of methods in different application scenarios and research directions in the future are given.
基金supported in part by the National Natural Science Foundation of China (NSFC) (Grant Nos. 62325221, 62132021, and 62372457), the Young Elite Scientists Sponsorship Program by CAST, China (Grant No. 2023QNRC001)the Natural Science Foundation of Hunan Province, China (Grant Nos. 2021RC3071 and 2022RC1104)the National University of Defense Technology Research Grants, China (Grant No. ZK22-52).
文摘Reconstructing from multi-view images is a longstanding problem in 3D vision,where neural radiance fields(NeRFs)have shown great potential and get realistic rendered images of novel views.Currently,most NeRF methods either require accurate camera poses or a large number of input images,or even both.Reconstructing NeRF from few-view images without poses is challenging and highly ill-posed.To address this problem,we propose CAD-NeRF,a method reconstructed from less than 10 images without any known poses.Specifically,we build a mini library of several CAD models from ShapeNet and render them from many random views.Given sparse-view input images,we run a model and pose retrieval from the library,to get a model with similar shapes,serving as the density supervision and pose initializations.Here we propose a multi-view pose retrieval method to avoid pose conflicts among views,which is a new and unseen problem in uncalibrated NeRF methods.Then,the geometry of the object is trained by the CAD guidance.The deformation of the density field and camera poses are optimized jointly.Then texture and density are trained and fine-tuned as well.All training phases are in self-supervised manners.Comprehensive evaluations of synthetic and real images show that CAD-NeRF successfully learns accurate densities with a large deformation from retrieved CAD models,showing the generalization abilities.
基金Project supported by the National Natural Science Foundation of China(Nos.61163016 and 61173125)
文摘Model classification is essential to the management and reuse of 3D CAD models.Manual model classification is laborious and error prone.At the same time,the automatic classification methods are scarce due to the intrinsic complexity of 3D CAD models.In this paper,we propose an automatic 3D CAD model classification approach based on deep neural networks.According to prior knowledge of the CAD domain,features are selected and extracted from 3D CAD models first,and then preprocessed as high dimensional input vectors for category recognition.By analogy with the thinking process of engineers,a deep neural network classifier for 3D CAD models is constructed with the aid of deep learning techniques.To obtain an optimal solution,multiple strategies are appropriately chosen and applied in the training phase,which makes our classifier achieve better performance.We demonstrate the efficiency and effectiveness of our approach through experiments on 3D CAD model datasets.
