Journal of Visual Communication and Image Representation最新文献

Rethinking the refinement stage of 3D object detection: A multi-task learning perspective with Mixture-of-Experts 重新思考三维目标检测的细化阶段：基于混合专家的多任务学习视角

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-06-01 Epub Date: 2026-05-25 DOI: 10.1016/j.jvcir.2026.104841

Bingqian Wu , Pei An , Siwen Quan , Qiao Wu , Linjie Li , Chu’ai Zhang , Jiaqi Yang

{"title":"Rethinking the refinement stage of 3D object detection: A multi-task learning perspective with Mixture-of-Experts","authors":"Bingqian Wu , Pei An , Siwen Quan , Qiao Wu , Linjie Li , Chu’ai Zhang , Jiaqi Yang","doi":"10.1016/j.jvcir.2026.104841","DOIUrl":"10.1016/j.jvcir.2026.104841","url":null,"abstract":"<div><div>Two-stage LiDAR-based 3D object detectors have achieved state-of-the-art accuracy, yet their performance is often limited by the refinement stage. In this work, we revisit 3D object refinement from a multi-task learning perspective and identify two independent sources of negative transfer: an <strong>inter-attribute conflict</strong>, where heterogeneous regression objectives, such as center, size, and orientation, interfere during joint optimization, and an <strong>inter-sample conflict</strong>, where proposals with varying point densities lead to gradient imbalance. To address these issues, we introduce two specialized Mixture-of-Experts architectures. The <strong>Attribute-MoE</strong> decouples regression objectives into dedicated expert branches to alleviate feature conflicts, while the <strong>Sparsity-MoE</strong> employs density-aware experts to adaptively refine proposals according to point sparsity. Integrated into strong two-stage baselines, our modules consistently improve performance on the KITTI dataset and the Waymo Open Dataset. Beyond empirical gains, our analysis reveals that Attribute-MoE and Sparsity-MoE solve largely independent problems, offering a practical “toolbox” for mitigating negative transfer in 3D object refinement and advancing adaptive, task-aware detector design. Code will be released at <span><span>https://github.com/12e21/RefineMoE</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":54755,"journal":{"name":"Journal of Visual Communication and Image Representation","volume":"118 ","pages":"Article 104841"},"PeriodicalIF":3.1,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178579","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}

引用次数: 0

Anchor node guided global–local graph neural networks for multimedia recommendation 锚节点导向的多媒体推荐全局-局部图神经网络

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-06-01 Epub Date: 2026-05-19 DOI: 10.1016/j.jvcir.2026.104835

Qi Ren, Desheng Cai

{"title":"Anchor node guided global–local graph neural networks for multimedia recommendation","authors":"Qi Ren, Desheng Cai","doi":"10.1016/j.jvcir.2026.104835","DOIUrl":"10.1016/j.jvcir.2026.104835","url":null,"abstract":"<div><div>Multimedia recommendation systems have garnered significant attention in recent years, particularly with GNN-based recommendation systems, which have demonstrated strong performance by aggregating information layer-by-layer through convolutional operations. However, previous studies have primarily focused on addressing the challenges of sparse interaction data, with limited attention given to issues arising from dense interactions. We argue that dense interactions hinder the accurate capture of user preferences, leading to bottleneck problem and an increased likelihood of noise. The introduction of GNNs further exacerbates these problems. Therefore, the problems arising from dense interaction cannot be ignored. In this paper, we propose a novel Anchor Node Guided Global–Local Graph Neural Networks for Multimedia Recommendation (AGG-LRec). To address the bottleneck problem caused by dense interactions, we propose a new graph neural network called anchor node module by using anchor nodes to aggregate the global information and interact with the target nodes. We also introduce two anchor-centric auxiliary objectives that explicitly regulate the semantic roles of anchor nodes. By enforcing anchor nodes to act as global semantic prototypes and maintaining cross-modal consistency at the anchor level, these objectives directly strengthen anchor-guided global aggregation under dense and noisy interactions. In addition, we propose a degree-sensitive edge pruning method for removing noise caused by dense interactions. This approach enhances the model’s robustness to noise by pruning edges. We also performed graph enhancement by introducing item–item edges. Experiments conducted on three datasets demonstrate that our proposed model outperforms baseline approaches. The source code is available at <span><span>https://github.com/ren3570/AGG-LRec</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":54755,"journal":{"name":"Journal of Visual Communication and Image Representation","volume":"118 ","pages":"Article 104835"},"PeriodicalIF":3.1,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178583","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}

引用次数: 0

Spatial position association-based 3D CT reconstruction from biplanar X-rays 基于空间位置关联的双平面x线三维CT重建

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-06-01 Epub Date: 2026-05-28 DOI: 10.1016/j.jvcir.2026.104845

Jiaze Li , Yang Li , Xueting Ren , Lijing Zhang , Yan Qiang , Juanjuan Zhao , Huajie Yue , Yan Wang

{"title":"Spatial position association-based 3D CT reconstruction from biplanar X-rays","authors":"Jiaze Li , Yang Li , Xueting Ren , Lijing Zhang , Yan Qiang , Juanjuan Zhao , Huajie Yue , Yan Wang","doi":"10.1016/j.jvcir.2026.104845","DOIUrl":"10.1016/j.jvcir.2026.104845","url":null,"abstract":"<div><div>Computed tomography is an advanced medical imaging technique that plays a crucial role in visualizing internal structures. However, it still faces challenges such as high radiation doses and expensive equipment costs. To reduce radiation exposure and improve imaging efficiency, this paper proposes a biplanar reconstruction model called SPAX2CT, which reconstructs high-quality 3D CT volumes from 2D X-ray images. The model incorporates a global frequency-domain fusion module and a local semantic enhancement module. These components strengthen global feature boundaries and compensate for the loss of semantic information in local features. Additionally, SPAX2CT employs spatial position association encoding to capture spatial relationships within the 3D structure, addressing the insufficient spatial position correlation found in traditional methods. Experimental results show that SPAX2CT significantly improves the modeling ability for complex spatial relationships and demonstrates excellent performance on the Lung Image Database Consortium. Our model outperforms baseline methods, while also showing strong scalability and applicability.</div></div>","PeriodicalId":54755,"journal":{"name":"Journal of Visual Communication and Image Representation","volume":"118 ","pages":"Article 104845"},"PeriodicalIF":3.1,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178586","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}

引用次数: 0

Landmark-assisted face and mouth 3D localization in smart spaces with a monocular camera 使用单目相机在智能空间中进行地标辅助面部和嘴部3D定位

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-06-01 Epub Date: 2026-05-26 DOI: 10.1016/j.jvcir.2026.104842

Frank Sanabria-Macias , Andrés Prados-Torreblanca , Marta Marron-Romera , Javier Macias-Guarasa , José M. Buenaposada , Luis Baumela , Sira E. Palazuelos-Cagigas

{"title":"Landmark-assisted face and mouth 3D localization in smart spaces with a monocular camera","authors":"Frank Sanabria-Macias , Andrés Prados-Torreblanca , Marta Marron-Romera , Javier Macias-Guarasa , José M. Buenaposada , Luis Baumela , Sira E. Palazuelos-Cagigas","doi":"10.1016/j.jvcir.2026.104842","DOIUrl":"10.1016/j.jvcir.2026.104842","url":null,"abstract":"<div><div>Accurate <span>3D</span> mouth localization from a single camera remains challenging in smart spaces, where speakers appear at low resolution and under unconstrained head poses. Most existing audiovisual tracking methods approximate the mouth position using a fixed location within the detected face bounding box, implicitly assuming frontal pose and leading to large errors under head rotations. In addition, monocular <span>3D</span> estimation suffers from scale ambiguity, i.e., the inherent uncertainty between object size and distance in <span>2D</span> image projections. We investigate whether head pose estimation (HPE) can overcome these limitations. Two monocular approaches are proposed: <span>LPoseVL</span>, combining facial landmarks with geometric pose recovery using a rigid <span>3D</span> head model, and <span>LFreeVL</span>, based on landmark-free 6DoF pose estimation with calibration. Experiments on the <span>AV16.3</span> and <span>CAV3D</span> datasets show that explicitly modeling head pose significantly improves accuracy, achieving up to 46.9% error reduction over bounding-box-based state-of-the-art methods.</div></div>","PeriodicalId":54755,"journal":{"name":"Journal of Visual Communication and Image Representation","volume":"118 ","pages":"Article 104842"},"PeriodicalIF":3.1,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148178585","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}

引用次数: 0

IDFreq: Identity-preserved human video generation via frequency-based decomposition IDFreq：通过基于频率的分解生成身份保留的人类视频

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-06-01 Epub Date: 2026-05-19 DOI: 10.1016/j.jvcir.2026.104828

Zhang Wan, Sheng Tang, Juan Cao, Yu Li

引用次数: 0

LWU-YOLO: A lightweight algorithm for small object detection in UAV applications LWU-YOLO：一种用于无人机小目标检测的轻量级算法

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-04-01 Epub Date: 2026-03-28 DOI: 10.1016/j.jvcir.2026.104791

Yapeng Li , Ting Wang , Tao Li , Xin Yang

引用次数: 0

GGCN: Gait Recognition with Generate Network and Convolutional Neural Network 基于生成网络和卷积神经网络的步态识别

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-04-01 Epub Date: 2026-03-27 DOI: 10.1016/j.jvcir.2026.104790

Hao Qin , Zhenxue Chen , Qingqiang Guo , Q.M. Jonathan Wu , Mengxu Lu

引用次数: 0

Beyond bounding boxes: Segmentation supervision for robust object detection in fisheye images 超越边界框：鱼眼图像中鲁棒目标检测的分割监督

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-04-01 Epub Date: 2026-04-04 DOI: 10.1016/j.jvcir.2026.104798

Arda Oztuner, Mehmet Kilicarslan

{"title":"Beyond bounding boxes: Segmentation supervision for robust object detection in fisheye images","authors":"Arda Oztuner, Mehmet Kilicarslan","doi":"10.1016/j.jvcir.2026.104798","DOIUrl":"10.1016/j.jvcir.2026.104798","url":null,"abstract":"<div><div>Fisheye cameras pose significant object detection challenges due to severe radial distortion, rendering traditional axis-aligned bounding boxes suboptimal for warped object shapes. We propose a pipeline that transforms bounding box annotations into instance segmentation masks using the Segment Anything Model (SAM) and validate mask fidelity against expert ground truth in both rectilinear and distorted domains. We benchmark various models on the Fisheye8K dataset, demonstrating the architectural generalizability of our approach across YOLOv8, YOLOv11, and YOLOv12. Results show that segmentation-based supervision yields substantial performance gains, improving the mean average precision (mAP@[0.5:0.95]) by up to 10 absolute points over models trained with bounding boxes, and up to 12 points in distorted outer regions. Furthermore, our approach outperforms state-of-the-art methods and establishes a new benchmark for fisheye object detection. This work highlights the specific theoretical and empirical benefits of automated segmentation-based annotation within complex, distorted imaging domains.</div></div>","PeriodicalId":54755,"journal":{"name":"Journal of Visual Communication and Image Representation","volume":"117 ","pages":"Article 104798"},"PeriodicalIF":3.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147656944","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}

引用次数: 0

Multi-view recursive gated convolutions for 3D object recognition and retrieval 三维目标识别与检索的多视图递归门控卷积

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-04-01 Epub Date: 2026-03-26 DOI: 10.1016/j.jvcir.2026.104792

Jiangzhong Cao, Yue Cai, Huan Zhang

{"title":"Multi-view recursive gated convolutions for 3D object recognition and retrieval","authors":"Jiangzhong Cao, Yue Cai, Huan Zhang","doi":"10.1016/j.jvcir.2026.104792","DOIUrl":"10.1016/j.jvcir.2026.104792","url":null,"abstract":"<div><div>Multi-view-based 3D shape recognition methods perform 3D object recognition and retrieval by processing series of images from various angles to generate a compact 3D descriptor. However, existing approaches often focus on integrating information from multiple views without addressing spatial interactions and redundancy when similar views are used. To overcome these challenges, we propose a novel framework, Multi-view Recursive Gated Convolutions (MVRGC). Our method first extracts features from multiple views at different scales, allowing for initial interaction of information across these views. Recursive gated convolutions are then applied to capture deeper spatial reciprocity and fine-tune feature interactions among views. Additionally, a preferred view module is introduced to reduce view redundancy by favoring distinctive and representative views. This module selects a subset of views that best describe the object while minimizing overlap. Experimental results on shape benchmark datasets demonstrate that MVRGC outperforms existing methods in 3D object recognition and retrieval tasks.</div></div>","PeriodicalId":54755,"journal":{"name":"Journal of Visual Communication and Image Representation","volume":"117 ","pages":"Article 104792"},"PeriodicalIF":3.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147600946","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}

引用次数: 0

Densely aggregated U-net with spatial-spectral interaction transformer for hyperspectral compressed imaging reconstruction 基于空间-光谱相互作用转换器的密集聚集u网用于高光谱压缩成像重建

IF 3.1 4区计算机科学

Journal of Visual Communication and Image Representation Pub Date : 2026-04-01 Epub Date: 2026-03-27 DOI: 10.1016/j.jvcir.2026.104795

Yun-Hui Li

{"title":"Densely aggregated U-net with spatial-spectral interaction transformer for hyperspectral compressed imaging reconstruction","authors":"Yun-Hui Li","doi":"10.1016/j.jvcir.2026.104795","DOIUrl":"10.1016/j.jvcir.2026.104795","url":null,"abstract":"<div><div>Hyperspectral imaging offers critical spectral information for applications such as material analysis and camouflage recognition. However, the acquisition of hyperspectral data cubes is inherently constrained by the Nyquist sampling theorem. While compressed sensing theory enables snapshot imaging by compressing the data cube into a 2D measurement, the ill-posed reconstruction remains a significant challenge. Recent deep learning methods, particularly vision transformers, have advanced the state-of-the-art (SOTA). Despite this, existing networks typically employ spectral or spatial self-attentions in isolation, blindly pursuing a global receptive field at the cost of computational efficiency and representational flexibility. Additionally, the vanilla skip connection in U-Nets is insufficient for effective multi-scale information transmission between encoder and decoder. To address these issues, we propose a Densely aggregated U-Net with a Spatial-Spectral Interaction Transformer (DSST). DSST parallelizes patch-based spectral self-attention and window-based spatial self-attention, complemented by an interaction mechanism. Furthermore, it introduces a densely aggregated skip connection to collect multi-scale features and bridge the semantic gap. Experimental results on both simulated and real-world scenes demonstrate that DSST achieves competitive performance with lower computational and memory costs compared to other end-to-end networks. Moreover, it offers faster inference speeds than deep unfolding networks.</div></div>","PeriodicalId":54755,"journal":{"name":"Journal of Visual Communication and Image Representation","volume":"117 ","pages":"Article 104795"},"PeriodicalIF":3.1,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147600948","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}

引用次数: 0