检索结果-内蒙古大学图书馆

31st International conference on Multimedia Modeling, MMM 2025

ISBN: (纸本)9789819620739

The proceedings contain 176 papers. The special focus in this conference is on Multimedia Modeling. The topics include: GWUNet: A UNet with Gated Attention and Improved Wavelet Transform for Thyroid Nodules Segmentation;HCV: Lightweight Hybrid CNN-Vision Transformer for Visual Object Tracking;hierArtEx: Hierarchical Representations and Art Experts Supporting the Retrieval of Museums in the Metaverse;hybrid Scalable Video Coding with Neural Compression and Enhancement for Streaming Media;Hyper-NeuS: Hypernetworks for Neural SdF Implicit Surface reconstruction by Volume Rendering;Image-Generation AI Model Retrieval by Contrastive Learning-Based Style distance Calculation;Improving Singing Voice Transcription Generalization with AI Generated Accompaniments;infrared Small Target detection with Feature Refinement and Context Enhancement;innovative Lifelog visualization and Exploration in virtual Reality – A Comparative Study;integrating S1 &S2 Framework for Enhanced Semantic Match in Person Re-identification;intra-class Compact Facial Expression Recognition Based on Amplitude Phase Separation;joint decision Network with Modality-Specific and dual Interactive Features for Fake News detection;kiite World: Socializing Map-Based Music Exploration Through Playlist Sharing and Synchronized Listening;kuzushijidiffuser: Japanese Kuzushiji Font Generation with Fontdiffuser;LIESA: Low-Light Image Enhancement with Semantic Awareness;lightweight dual Grouped Large-Kernel Convolutions for Salient Object detection Network;lightweight Motion-Aware Video Super-Resolution for Compressed Videos;LITA: LMM-Guided Image-Text Alignment for Art Assessment;LLMs-Based Augmentation for domain Adaptation in Long-Tailed Food datasets;making Strides Security in Multimodal Fake News detection Models: A Comprehensive Analysis of Adversarial Attacks;mambaTalk: Speech-driven 3d Facial Animation with Mamba;MC-YOLO: Multi-scale Transmission Line defect Target Recognition Network;MdT-Net: A Mask decoder Tu

关键词：

来源：评论

学校读者我要写书评

暂无评论

A web application with multi-input capabilities for AI-driven 3d object generation, designed for XR applications

A web application with multi-input capabilities for AI-drive...

引用

2025 IEEE conference on virtual Reality and 3d User Interfaces Abstracts and Workshops, VRW 2025

作者： Kawka, Monika Sara Nikopoulos, Alexandros Bikiris, Panagiotis Kalisperakis, Ilias Stentoumis, Christos up2metric Greece

ISBN: (纸本)9798331514846

The generation of 3d content is essential in many applications and especially the ones relying on eXtended Reality (XR). However, creating XR-ready models is typically a costly and timely manual process, as well as a complex task for non-expert users. Here, a platform for multi-input 3d model generation is presented. It integrates, in a single point, methods to generate XR-ready 3d objects from three input types: multiple images, a single image, or a text prompt. Each input is managed by a separate service, offered to users either via an API, to support integration with other services, or via a dedicated intuitive web-application, designed for non-experts. The web application contains tools that ease the usage of developed services and allow scheduling operations, viewing results, and post-processing of 3d models in a user-friendly manner. It also provides additional features, such as simplification, remeshing, and levels of detail generation, developed to make the generated models even more suitable for XR applications. © 2025 IEEE.

关键词： 3d object generation 3d reconstruction AI-based 3d modeling XR applications

来源：评论

学校读者我要写书评

暂无评论

Using Signed distance Fields to Achieve Temporal Compression of Mesh-Based Volumetric Video

Using Signed Distance Fields to Achieve Temporal Compression...

引用

2025 IEEE conference on virtual Reality and 3d User Interfaces Abstracts and Workshops, VRW 2025

作者： Anderson, Jakob E. Li, Yaojie Hogue, Andrew Gamer Lab Ontario Tech University Oshawa ON Canada

ISBN: (纸本)9798331514846

Volumetric Video is an exciting medium that enables visualization of 3d time-varying data, often as sequence of Point Clouds or Textured Meshes. despite over a decade of focus, there still exist several problems that make it challenging to employ. This paper focuses on one aspect of deploying Volumetric Video - data compression. Reducing the data storage size is necessary for streaming applications due to the large size of sequences consisting of even a few seconds of footage. This work explores the use of temporal compression of time-varying textured meshes using Signed distance Fields and compare with the state-of-the art techniques. The goal is to produce a novel compression method to maximize a size-to-error compression ratio while maintaining a decompression speed of 30 frames per second. The results reflect that as file sizes decrease, the proposed method's error increases at a slower rate than the state-of-the-art. © 2025 IEEE.

关键词： Compression Signed distance Field Surface reconstruction Time Varying Mesh Volumetric Video

来源：评论

学校读者我要写书评

暂无评论

3d reconstruction Method of Indoor Scene Layout based on Neural Network

3D Reconstruction Method of Indoor Scene Layout based on Neu...

引用

Intelligent Systems and Computational Networks (ICISCN), International conference on

作者： Yingying Lin Changchun University of Architecture and Civil Engineering Changchun China

ISBN: (数字)9798331529246

ISBN: (纸本)9798331529253

In recent years, three-dimensional (3d) reconstruction technology of indoor scenes has been widely used in fields such as virtual reality and indoor navigation, but existing methods still have shortcomings in complex layout and occlusion processing. First, this paper uses a scene segmentation method based on convolutional neural network (CNN) to segment the input indoor image data into layers and extract feature information at different levels such as room structure, furniture, and objects. By training with a large amount of labeled data, the network can precisely identify different objects and distinguish their positions and boundaries in the scene. As the segmentation fineness increases (from 1 to 5), the mean depth error (MdE) decreases from 0.14 to 0.07. The structural similarity index (SSIM) gradually increases from 0.72 to 0.92. Through effective processing of different depths and scales, a new solution is provided for the 3d modeling of future indoor scenes.

关键词： Training Solid modeling Adaptation models Image segmentation Three-dimensional displays Computational modeling Neural networks Layout Feature extraction Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Semantics-Controlled Gaussian Splatting for Outdoor Scene reconstruction and Rendering in virtual Reality

Semantics-Controlled Gaussian Splatting for Outdoor Scene Re...

引用

IEEE Annual International Symposium virtual Reality

作者： Hannah Schieber Jacob Young Tobias Langlotz Stefanie Zollmann daniel Roth Technical University of Munich Human-Centered Computing and Extended Reality Lab TUM University Hospital Clinic for Orthopedics and Sports Orthopedics Munich Germany Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU) Erlangen Germany Department of Computer Science University of Otago Dunedin New Zealand Department of Computer Science Aarhus University Aarhus Denmark

ISBN: (数字)9798331536459

ISBN: (纸本)9798331536466

Advancements in 3d rendering like Gaussian Splatting (GS) allow novel view synthesis and real-time rendering in virtual reality (VR). However, GS-created 3d environments are often difficult to edit. For scene enhancement or to incorporate 3d assets, segmenting Gaussians by class is essential. Existing segmentation approaches are typically limited to certain types of scenes, e.g., "circular" scenes, to determine clear object boundaries. However, this method is ineffective when removing large objects in non-"circling" scenes such as large outdoor *** propose Semantics-Controlled GS (SCGS), a segmentation-driven GS approach, enabling the separation of large scene parts in uncontrolled, natural environments. SCGS allows scene editing and the extraction of scene parts for VR. Additionally, we introduce a challenging outdoor dataset, overcoming the "circling" setup. We outperform the state-of-the-art in visual quality on our dataset and in segmentation quality on the 3d-OVS dataset. We conducted an exploratory user study, comparing a 360-video, plain GS, and SCGS in VR with a fixed viewpoint. In our subsequent main study, users were allowed to move freely, evaluating plain GS and SCGS. Our main study results show that participants clearly prefer SCGS over plain GS. We overall present an innovative approach that surpasses the state-of-the-art both technically and in user experience.

关键词： visualization Three-dimensional displays Semantics virtual reality User interfaces Rendering (computer graphics) User experience Real-time systems

来源：评论

学校读者我要写书评

暂无评论

LLM-powered Gaussian Splatting in VR interactions

LLM-powered Gaussian Splatting in VR interactions

引用

virtual Reality and 3d User Interfaces Abstracts and Workshops (VRW), IEEE conference on

作者： Haotian Mao Zhuoxiong Xu Siyue Wei Yule Quan Nianchen deng Xubo Yang Shanghai Jiao Tong University Shanghai AI Lab

ISBN: (数字)9798331514846

ISBN: (纸本)9798331525637

Recent advances in radiance field rendering, particularly 3d Gaussian Splatting (3dGS), have demonstrated significant potential for VR content creation, offering both high-quality rendering and an efficient production pipeline. However, current physics-based interaction systems for 3dGS are limited to either simplistic, unrealistic simulations or require substantial user input for complex scenes, largely due to the lack of scene comprehension. In this demonstration, we present a highly realistic interactive VR system powered by large language models (LLMs). After object-aware GS reconstruction, we prompt GPT-4o to analyze the physical properties of objects in the scene, which then guide physical simulations that adhere to real-world phenomena. Additionally, We design a GPT-assisted GS inpainting module to complete the areas occluded by manipulated objects. To facilitate rich interaction, we introduce a computationally efficient physical simulation framework through a PBd-based unified interpolation method, which supports various forms of physical interactions. In our research demonstrations, we reconstruct varieties of scenes enhanced by LLM’s understanding, showcasing how our VR system can support complex, realistic interactions without additional manual design or annotation.

关键词： Solid modeling Interpolation Three-dimensional displays Computational modeling Large language models Pipelines virtual reality Production User interfaces Rendering (computer graphics)

来源：评论

学校读者我要写书评

暂无评论

A survey of multi-view stereo 3d reconstruction algorithms based on deep learning

引用

digital Signal Processing: A Review Journal 2025年 165卷

作者： Feng, Shengjie Wu, Xiaoqun Cao, Jian School of Computer and Artificial Intelligence Beijing Technology and Business University Beijing100048 China

Multi-view stereo (MVS) 3d surface reconstruction, as a core problem in the fields of computer vision and graphics, aims to accurately recover the geometric structure of a scene from multi-view images. This technology bridges the gap between data capture and surface editing. It can be used in various downstream applications such as cultural heritage preservation, urban construction planning, virtual Reality (VR), and Augmented Reality (AR). However, MVS still faces numerous challenges when dealing with complex scenes, high-frequency textures, and occlusions. Improving accuracy, robustness, and computational efficiency has become a key focus of current research. To address the aforementioned challenges, this survey systematically reviews key techniques in MVS-based 3d reconstruction, focusing on the latest advancements in deep learning methods. It provides a detailed analysis of the general MVS pipeline, including feature extraction, cost volume construction, cost volume regularization, and loss functions. Additionally, we compile commonly used datasets for MVS 3d reconstruction (with corresponding links) and evaluation metrics. In response to the performance limitations of traditional methods in complex scenes, we analyze three representative paradigms based on the evolution of network architectures: convolutional neural networks (CNNs), recurrent neural networks (RNNs), and Transformer-based structures. Furthermore, to broaden research perspectives, we explore and summarize the applications of emerging Neural Radiance Fields (NeRF) and 3d Gaussian Splatting (3dGS) technologies in MVS, emphasizing a comparative analysis of their implementation processes. For experimental evaluation, this paper reviews the performance of leading MVS approaches on several public datasets, and compares their effectiveness. Finally, we summarize the theoretical and practical significance of this survey, highlighting its main contributions, strengths, and limitations, proposing future re

关键词： 3d reconstruction

来源：评论

学校读者我要写书评

暂无评论

design of patient-specific mandibular reconstruction plates and a hybrid scaffold

引用

Computers in Biology and Medicine 2025年 184卷 109380-109380页

作者： dogan, Sait Emre Ozturk, Cengizhan Koc, Bahattin Bogazici University Institute of Biomedical Engineering Istanbul34684 Turkey 3D Bioprinting Laboratory Sabanci University Nanotechnology Research and Application Center Istanbul34956 Turkey Faculty of Engineering and Natural Sciences Sabanci University Istanbul34956 Turkey

Background: Managing segmental mandibular defects remains challenging, requiring a multidisciplinary approach despite the remarkable progress in mandibular reconstruction plates, finite element methods, computer-aided design and manufacturing techniques, and novel surgical procedures. complex surgeries require a comprehensive approach, as using only reconstruction plates or tissue scaffolds may not be adequate for optimal results. The limitations of the treatment options should be investigated towards a patient-specific trend to provide shorter surgery time, better healing, and lower costs. Integrated hybrid scaffold systems are promising in improving mechanical properties and facilitating healing. By combining different materials and structures, hybrid scaffolds can provide enhanced support and stability to the tissue regeneration process, leading to better patient outcomes. The use of such systems represents a significant advancement in tissue engineering and a wide range of medical procedures. Materials and Methods: A head and neck computed tomography (CT) data of a patient with odontogenic myxoma was used for creating a three-dimensional (3d) mandible model. virtual osteotomies were performed to create a segmental defect model, including the angulus mandibulae region. The first mandibular reconstruction plate was designed. Finite elemental analyses (FEA) and topology optimizations were performed to create two different reconstruction plates for different treatment scenarios. The FEA were performed for the resulting two plates to assess their biomechanical performance. To provide osteoconductive and osteoinductive properties a scaffold was designed using the defect area. A biomimetic Tricalcium phosphate-Polycaprolactone (TCP-PCL) hybrid bone scaffold enhanced with Hyaluronic acid dipping was manufactured. Results: The results of the in-silico analysis indicate that the designed reconstruction plates possess robust biomechanical performance and demonstrate remark

关键词： Hyaluronic acid

来源：评论

学校读者我要写书评

暂无评论

PVT: An Implicit Surface reconstruction Framework via Point Voxel Geometric-Aware Transformer

PVT: An Implicit Surface Reconstruction Framework via Point ...

引用

IEEE Workshop on Applications of Computer Vision (WACV)

作者： Chuanmao Fan Chenxi Zhao Ye duan University of Missouri-Columbia Columbia MO USA Clemson University Clemson SC USA

ISBN: (数字)9798331510831

ISBN: (纸本)9798331510848

3d surface reconstruction from unorganized point clouds is a fundamental task in visual computing with numerous applications in areas such as robotics, virtual reality, augmented reality, and animation. To date, many deep learning-based surface reconstruction methods have been proposed, demonstrating great performance on many benchmark datasets. Among these, neural implicit field learning-based methods have gained popularity for their capability of representing complex structures in a continuous implicit distance field. Existing neural implicit field learning methods either utilize voxelized point cloud then feed them to a deep network, or directly take points as input. In this paper, we propose an implicit surface reconstruction framework based on point voxel geometric-aware transformer PVT to seamlessly integrate point-based convolution with voxel-based convolution using bidirectional transformers. Experiments show that the proposed PVT framework can better encode local geometry details and provide a significant performance boost over existing state-of-the-art methods.

关键词： Learning systems Point cloud compression Geometry Surface reconstruction Three-dimensional displays Shape Convolution Benchmark testing reconstruction algorithms Transformers

来源：评论

学校读者我要写书评

暂无评论

GANESH: Generalizable NeRF for Lensless Imaging

GANESH: Generalizable NeRF for Lensless Imaging

引用

IEEE Workshop on Applications of Computer Vision (WACV)

作者： Rakesh Raj Madavan Akshat Kaimal Badhrinarayanan K V Vinayak Gupta Rohit Choudhary Chandrakala Shanmuganathan Kaushik Mitra Shiv Nadar University Chennai Indian Institute of Technology Madras

ISBN: (数字)9798331510831

ISBN: (纸本)9798331510848

Lensless imaging offers a significant opportunity to develop ultra-compact cameras by removing the conventional bulky lens system. However, without a focusing element, the sensor's output is no longer a direct image but a complex multiplexed scene representation. Traditional methods have attempted to address this challenge by employing learnable inversions and refinement models, but these methods are primarily designed for 2d reconstruction and do not generalize well to 3d reconstruction. We introduce GANESH, a novel framework designed to enable simultaneous refinement and novel view synthesis from multi-view lensless images. Unlike existing methods that require scene-specific training, our approach supports on-the-fiy inference without retraining on each scene. Moreover, our framework allows us to tune our model to specific scenes, enhancing the rendering and refinement quality. To facilitate research in this area, we also present the first multi-view lensless dataset, LenslessScenes. Extensive experiments demonstrate that our method outperforms current approaches in reconstruction accuracy and refinement quality. Code and video results are available here.

关键词： Training Solid modeling Three-dimensional displays virtual reality Rendering (computer graphics) Neural radiance field Robustness Wearable devices Image reconstruction Synthetic data

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：