Computer Vision and Image Understanding最新文献_第3页

A comprehensive review on advances in instance-level 6D object pose tracking 实例级6D物体姿态跟踪研究进展综述

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-02-01 DOI: 10.1016/j.cviu.2026.104667

Yanming Wu , Hui Zhang , Patrick Vandewalle , Peter Slaets , Eric Demeester

Instance-level 6D object pose tracking involves tracking a known object in 3D space and estimating its six degrees of freedom (6DoF) pose across consecutive images, starting from the initial pose in the first frame. This technology has wide-ranging applications in various fields, including robotics, augmented reality, and human–machine interaction. Over the years, significant progress has been made in this field. Many methods tackle the problem of instance-level 6D pose tracking from RGB images. These techniques can be classified based on their use of keypoints, edges, region information or direct optimization. Additionally, the availability of affordable RGB-D sensors has prompted the utilization of depth data for 6D pose tracking. Another notable advancement is the adoption of deep neural networks, which have shown promising results. Despite these developments, survey studies on the latest advancements in this field are lacking. Therefore, this work aims to fill this gap by providing a comprehensive review of recent progress in instance-level 6D object tracking, covering the aforementioned advancements. This paper provides a detailed examination of metrics, datasets, and methodology employed in this field. Based on the problem modeling approach, methods reviewed in this paper are categorized into optimization-based, learning-based, filtering-based approaches and hybrid approaches that combine various techniques. Furthermore, quantitative results on several publicly available datasets are presented and analyzed, along with applications and open challenges for future research directions.

实例级6D对象姿态跟踪涉及跟踪3D空间中的已知对象，并从第一帧中的初始姿态开始，在连续图像中估计其六个自由度（6DoF）姿态。该技术在机器人、增强现实、人机交互等领域有着广泛的应用。多年来，这一领域取得了重大进展。许多方法解决了RGB图像的实例级6D姿态跟踪问题。这些技术可以根据关键点、边缘、区域信息或直接优化的使用进行分类。此外，价格实惠的RGB-D传感器的可用性促使深度数据用于6D姿态跟踪。另一个显著的进步是深度神经网络的采用，它已经显示出有希望的结果。尽管有这些发展，但缺乏对这一领域最新进展的调查研究。因此，这项工作旨在通过提供实例级6D对象跟踪的最新进展的全面审查来填补这一空白，涵盖上述进展。本文提供了一个详细的检查指标，数据集，并在该领域采用的方法。基于问题建模方法，本文综述的方法可分为基于优化的方法、基于学习的方法、基于过滤的方法和结合各种技术的混合方法。此外，对几个公开可用数据集的定量结果进行了介绍和分析，以及未来研究方向的应用和开放挑战。

{"title":"A comprehensive review on advances in instance-level 6D object pose tracking","authors":"Yanming Wu , Hui Zhang , Patrick Vandewalle , Peter Slaets , Eric Demeester","doi":"10.1016/j.cviu.2026.104667","DOIUrl":"10.1016/j.cviu.2026.104667","url":null,"abstract":"<div><div>Instance-level 6D object pose tracking involves tracking a known object in 3D space and estimating its six degrees of freedom (6DoF) pose across consecutive images, starting from the initial pose in the first frame. This technology has wide-ranging applications in various fields, including robotics, augmented reality, and human–machine interaction. Over the years, significant progress has been made in this field. Many methods tackle the problem of instance-level 6D pose tracking from RGB images. These techniques can be classified based on their use of keypoints, edges, region information or direct optimization. Additionally, the availability of affordable RGB-D sensors has prompted the utilization of depth data for 6D pose tracking. Another notable advancement is the adoption of deep neural networks, which have shown promising results. Despite these developments, survey studies on the latest advancements in this field are lacking. Therefore, this work aims to fill this gap by providing a comprehensive review of recent progress in instance-level 6D object tracking, covering the aforementioned advancements. This paper provides a detailed examination of metrics, datasets, and methodology employed in this field. Based on the problem modeling approach, methods reviewed in this paper are categorized into optimization-based, learning-based, filtering-based approaches and hybrid approaches that combine various techniques. Furthermore, quantitative results on several publicly available datasets are presented and analyzed, along with applications and open challenges for future research directions.</div></div>","PeriodicalId":50633,"journal":{"name":"Computer Vision and Image Understanding","volume":"264 ","pages":"Article 104667"},"PeriodicalIF":3.5,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146078430","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Interaction-aware representation learning for action quality assessment in freestyle skiing big air 基于交互感知表征学习的自由式滑雪大空中动作质量评估

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-02-01 DOI: 10.1016/j.cviu.2026.104634

Shiyue Chen , Yanchao Liu , Ziyue Wang , Xina Cheng , Takeshi Ikenaga

Freestyle skiing big air requires precise athlete–ski coordination to determine both technical difficulty and execution quality. Accurate action quality assessment in this discipline therefore necessitates explicit modeling of human–object interactions. However, most existing methods rely on video-level or human-centric representations, overlooking structured athlete-ski relationships and limiting evaluation of control and stability. To address this, we construct a freestyle skiing big air dataset with fine-grained annotations, including frame-level athlete-ski bounding boxes and performance-related metadata. Based on this dataset, we propose an interaction-aware framework that captures athlete–ski coordination by combining instance-level appearance and positional features through spatiotemporal reasoning. Furthermore, to avoid commonly used uniform sampling diluting performance-critical moments in long sequences, we introduce a training-free entropy-based sampling strategy that exploits athlete–ski geometric dynamics to identify performance-critical moments such as take-off, rotation, and landing, thereby reducing redundancy. Together, these designs address where to look and when to focus in big air assessment. Extensive experiments demonstrate that our method achieves a Spearman’s rank correlation of 0.7173 on the proposed dataset, outperforming state-of-the-art methods.

自由式滑雪大空中需要精确的运动员-滑雪板协调，以确定技术难度和执行质量。因此，在这一学科中，准确的行动质量评估需要对人与物之间的相互作用进行明确的建模。然而，大多数现有方法依赖于视频级或以人为中心的表示，忽略了结构化的运动员-滑雪关系，限制了对控制和稳定性的评估。为了解决这个问题，我们构建了一个带有细粒度注释的自由式滑雪大空气数据集，包括帧级运动员滑雪边界框和与性能相关的元数据。基于此数据集，我们提出了一个交互感知框架，该框架通过时空推理结合实例级外观和位置特征来捕获运动员-滑雪协调。此外，为了避免通常使用的均匀采样在长序列中稀释性能关键时刻，我们引入了一种基于无训练熵的采样策略，该策略利用运动员滑雪的几何动力学来识别起飞、旋转和着陆等性能关键时刻，从而减少冗余。总之，这些设计说明了在大空气评估中应该关注哪里以及何时关注。大量的实验表明，我们的方法在提出的数据集上实现了0.7173的Spearman秩相关，优于最先进的方法。

{"title":"Interaction-aware representation learning for action quality assessment in freestyle skiing big air","authors":"Shiyue Chen , Yanchao Liu , Ziyue Wang , Xina Cheng , Takeshi Ikenaga","doi":"10.1016/j.cviu.2026.104634","DOIUrl":"10.1016/j.cviu.2026.104634","url":null,"abstract":"<div><div>Freestyle skiing big air requires precise athlete–ski coordination to determine both technical difficulty and execution quality. Accurate action quality assessment in this discipline therefore necessitates explicit modeling of human–object interactions. However, most existing methods rely on video-level or human-centric representations, overlooking structured athlete-ski relationships and limiting evaluation of control and stability. To address this, we construct a freestyle skiing big air dataset with fine-grained annotations, including frame-level athlete-ski bounding boxes and performance-related metadata. Based on this dataset, we propose an interaction-aware framework that captures athlete–ski coordination by combining instance-level appearance and positional features through spatiotemporal reasoning. Furthermore, to avoid commonly used uniform sampling diluting performance-critical moments in long sequences, we introduce a training-free entropy-based sampling strategy that exploits athlete–ski geometric dynamics to identify performance-critical moments such as take-off, rotation, and landing, thereby reducing redundancy. Together, these designs address <em>where to look</em> and <em>when to focus</em> in big air assessment. Extensive experiments demonstrate that our method achieves a Spearman’s rank correlation of 0.7173 on the proposed dataset, outperforming state-of-the-art methods.</div></div>","PeriodicalId":50633,"journal":{"name":"Computer Vision and Image Understanding","volume":"264 ","pages":"Article 104634"},"PeriodicalIF":3.5,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146078741","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

3D-aware virtual try-on using only 2D inputs 仅使用2D输入的3d感知虚拟试戴

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-02-01 DOI: 10.1016/j.cviu.2026.104661

Jaeyoon Lee , Hojoon Jung , Jongwon Choi

We present 3DFit, which is a novel 3D-aware virtual try-on framework that synthesizes realistic try-on images using only 2D inputs. Unlike previous methods that either ignore 3D body geometry or rely entirely on 3D clothing models, 3DFit utilizes 3D human meshes estimated from 2D images and adaptively transforms 3D clothing templates guided by 2D clothing images. We further introduce a warping strategy that integrates 3D information into 2D clothing images using a set of pre-designed 3D templates, enabling efficient adaptation to various body shapes and poses. As a result, our method supports accurate and personalized virtual try-on experiences. Experimental results on the VITON-HD dataset demonstrate that 3DFit outperforms existing methods in preserving garment structure and maintaining high visual quality across a wide range of body types and poses.

我们提出3DFit，这是一种新颖的3d感知虚拟试戴框架，仅使用2D输入合成逼真的试戴图像。与以往的方法忽略3D人体几何或完全依赖3D服装模型不同，3DFit利用从2D图像估计的3D人体网格，并在2D服装图像的指导下自适应变换3D服装模板。我们进一步介绍了一种翘曲策略，该策略使用一组预先设计的3D模板将3D信息集成到2D服装图像中，从而能够有效地适应各种体型和姿势。因此，我们的方法支持准确和个性化的虚拟试戴体验。VITON-HD数据集上的实验结果表明，3DFit在保留服装结构和保持高视觉质量方面优于现有方法，适用于各种体型和姿势。

引用次数: 0

FusGaze: Full range gaze estimation with multi-scale fusion FusGaze：全范围凝视估计与多尺度融合

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-01-30 DOI: 10.1016/j.cviu.2026.104666

Dayeon Yoo, Jaehun Cho, Kwangho Song

Gaze estimation plays a crucial role in a wide range of applications, including human–computer interaction (HCI), augmented and virtual reality (AR/VR), driver monitoring and robotic intelligence. Most of the conventional approaches rely on external detectors to extract head or face regions prior to gaze estimation. It causes a fundamental issue by inheriting the detector errors, which leads to incorrect gaze predictions based on inaccurate input data. To overcome these limitations, we propose FusGaze, a novel end-to-end gaze estimation framework that seamlessly integrates detection and estimation within a single architecture. FusGaze eliminates external detector dependencies by appending an object detection head and a gaze estimation head to a single detection backbone, thereby enabling more stable and robust gaze estimation in practical environments. To enable effective multi-scale fusion, we design an Adaptive Weighted Late Fusion (AWLF) block that data-drivenly fuses multi-scale gaze vectors. Moreover, a 2D Gaussian heatmap-based auxiliary loss function is introduced to improve estimation stability, even under challenging scenarios such as side or rear views where visual cues are limited. Experiments conducted on the Gaze360 benchmark demonstrate that FusGaze achieves state-of-the-art performance with mean angular error of 12.85°, among single image-based gaze estimation models. In particular, the model shows significant improvements under front-facing conditions while maintaining competitive performance under backward-facing conditions, thus experimentally validating its effectiveness in full 360° gaze estimation.

注视估计在广泛的应用中起着至关重要的作用，包括人机交互（HCI），增强和虚拟现实（AR/VR），驾驶员监控和机器人智能。大多数传统方法依赖于外部检测器来提取头部或面部区域，然后再进行凝视估计。它通过继承检测器错误导致了一个根本性的问题，这导致基于不准确的输入数据的不正确的凝视预测。为了克服这些限制，我们提出了一种新的端到端凝视估计框架FusGaze，它将检测和估计无缝集成在一个架构中。FusGaze通过将目标检测头和凝视估计头附加到单个检测骨干来消除外部检测器依赖，从而在实际环境中实现更稳定和鲁棒的凝视估计。为了实现有效的多尺度融合，我们设计了一个自适应加权后期融合（AWLF）块，该块数据驱动地融合了多尺度凝视向量。此外，引入了一个基于二维高斯热图的辅助损失函数，以提高估计的稳定性，即使在具有挑战性的场景下，如视觉线索有限的侧视图或后视图。在Gaze360基准测试上进行的实验表明，在基于单图像的凝视估计模型中，FusGaze的平均角度误差为12.85°，达到了最先进的性能。特别是，该模型在正面条件下显示出显着的改进，同时在背面条件下保持竞争性能，从而通过实验验证了其在全360°凝视估计中的有效性。

{"title":"FusGaze: Full range gaze estimation with multi-scale fusion","authors":"Dayeon Yoo, Jaehun Cho, Kwangho Song","doi":"10.1016/j.cviu.2026.104666","DOIUrl":"10.1016/j.cviu.2026.104666","url":null,"abstract":"<div><div>Gaze estimation plays a crucial role in a wide range of applications, including human–computer interaction (HCI), augmented and virtual reality (AR/VR), driver monitoring and robotic intelligence. Most of the conventional approaches rely on external detectors to extract head or face regions prior to gaze estimation. It causes a fundamental issue by inheriting the detector errors, which leads to incorrect gaze predictions based on inaccurate input data. To overcome these limitations, we propose FusGaze, a novel end-to-end gaze estimation framework that seamlessly integrates detection and estimation within a single architecture. FusGaze eliminates external detector dependencies by appending an object detection head and a gaze estimation head to a single detection backbone, thereby enabling more stable and robust gaze estimation in practical environments. To enable effective multi-scale fusion, we design an Adaptive Weighted Late Fusion (AWLF) block that data-drivenly fuses multi-scale gaze vectors. Moreover, a 2D Gaussian heatmap-based auxiliary loss function is introduced to improve estimation stability, even under challenging scenarios such as side or rear views where visual cues are limited. Experiments conducted on the Gaze360 benchmark demonstrate that FusGaze achieves state-of-the-art performance with mean angular error of 12.85°, among single image-based gaze estimation models. In particular, the model shows significant improvements under front-facing conditions while maintaining competitive performance under backward-facing conditions, thus experimentally validating its effectiveness in full 360° gaze estimation.</div></div>","PeriodicalId":50633,"journal":{"name":"Computer Vision and Image Understanding","volume":"265 ","pages":"Article 104666"},"PeriodicalIF":3.5,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191752","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

FWS-Net: Fixed-Point Deep Unfolding Network for compressed sensing via multi-scale wavelet fusion and state-space modeling FWS-Net：基于多尺度小波融合和状态空间建模的压缩感知定点深度展开网络

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-01-29 DOI: 10.1016/j.cviu.2025.104630

Hongbing Wu , Siyang Huang , Xiaojun Guo , Xiaoyong Sun , Bei Sun , Zhuo Zhang , Shaojing Su

Deep Unfolding Networks (DUNs) have emerged as a prominent direction in Compressed Sensing (CS) due to their interpretability and superior reconstruction performance. However, as image reconstruction tasks become increasingly complex, existing DUNs face two critical challenges: (i) Balancing the extraction of local details and global structure while reducing computational complexity; (ii) Ignoring frequency domain information in cross-stage feature fusion, leading to feature loss, artifact accumulation, and structural distortion. To address these challenges, we propose FWS-Net—a Fixed-Point Deep Unfolding Network that integrates multi-scale wavelet fusion and state-space modeling. Our method is theoretically grounded in the Fixed-Point Continuation (FPC) algorithm, enabling physically consistent reconstruction through adaptive soft-thresholding. Specifically, each stage incorporates a Wavelet State Space Reconstruction (WSSR) module to jointly optimize multi-scale feature representation and extend the sparse proximal projection, enhancing cross-stage information interaction. Within each stage, a Dual-Branch State Space (DBSS) module facilitates collaborative recovery of local details and global consistency. Across stages, an Inter-Stage Wavelet Fusion (ISWF) mechanism disentangles and exchanges high and low frequency components, alleviating feature distortion and information discontinuity. Extensive experiments on image CS and CS-MRI demonstrate that FWS-Net outperforms state-of-the-art methods while retaining high explainability and computational efficiency.

深度展开网络（DUNs）由于其可解释性和优越的重构性能而成为压缩感知（CS）研究的一个重要方向。然而，随着图像重建任务的日益复杂，现有的DUNs面临两个关键挑战：(i)在降低计算复杂度的同时平衡局部细节和全局结构的提取；（ii）在跨阶段特征融合中忽略频域信息，导致特征丢失、伪影积累、结构失真。为了解决这些挑战，我们提出了fws - net -一个集成了多尺度小波融合和状态空间建模的定点深度展开网络。我们的方法在理论上基于不动点延拓（FPC）算法，通过自适应软阈值实现物理一致的重建。具体而言，每个阶段都包含一个小波状态空间重构（WSSR）模块，共同优化多尺度特征表示并扩展稀疏近端投影，增强了跨阶段的信息交互。在每个阶段中，双分支状态空间（DBSS）模块有助于协作恢复本地细节和全局一致性。在各个阶段，一种阶段间小波融合（ISWF）机制分离和交换高低频分量，减轻特征失真和信息不连续。在图像CS和CS- mri上进行的大量实验表明，FWS-Net在保持高可解释性和计算效率的同时，优于最先进的方法。

{"title":"FWS-Net: Fixed-Point Deep Unfolding Network for compressed sensing via multi-scale wavelet fusion and state-space modeling","authors":"Hongbing Wu , Siyang Huang , Xiaojun Guo , Xiaoyong Sun , Bei Sun , Zhuo Zhang , Shaojing Su","doi":"10.1016/j.cviu.2025.104630","DOIUrl":"10.1016/j.cviu.2025.104630","url":null,"abstract":"<div><div>Deep Unfolding Networks (DUNs) have emerged as a prominent direction in Compressed Sensing (CS) due to their interpretability and superior reconstruction performance. However, as image reconstruction tasks become increasingly complex, existing DUNs face two critical challenges: (i) Balancing the extraction of local details and global structure while reducing computational complexity; (ii) Ignoring frequency domain information in cross-stage feature fusion, leading to feature loss, artifact accumulation, and structural distortion. To address these challenges, we propose FWS-Net—a Fixed-Point Deep Unfolding Network that integrates multi-scale wavelet fusion and state-space modeling. Our method is theoretically grounded in the Fixed-Point Continuation (FPC) algorithm, enabling physically consistent reconstruction through adaptive soft-thresholding. Specifically, each stage incorporates a Wavelet State Space Reconstruction (WSSR) module to jointly optimize multi-scale feature representation and extend the sparse proximal projection, enhancing cross-stage information interaction. Within each stage, a Dual-Branch State Space (DBSS) module facilitates collaborative recovery of local details and global consistency. Across stages, an Inter-Stage Wavelet Fusion (ISWF) mechanism disentangles and exchanges high and low frequency components, alleviating feature distortion and information discontinuity. Extensive experiments on image CS and CS-MRI demonstrate that FWS-Net outperforms state-of-the-art methods while retaining high explainability and computational efficiency.</div></div>","PeriodicalId":50633,"journal":{"name":"Computer Vision and Image Understanding","volume":"265 ","pages":"Article 104630"},"PeriodicalIF":3.5,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191765","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

CPNText-DETR: Dynamic contour initialization and normalized loss for accurate arbitrarily shaped text detection CPNText-DETR：用于精确任意形状文本检测的动态轮廓初始化和归一化损失

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-01-29 DOI: 10.1016/j.cviu.2026.104647

Yang Wang , Wanlin Zhou , Chen Lv , Hongjun Xia , Jiakai Zhou

Recent transformer-based methods for arbitrarily shaped text detection have shown improved performance by predicting text contour points. However, these methods often rely on bounding box centers for text contour points localization, which may not accurately fall within the text area, and use L1 loss for contour point regression, causing imbalances across scales. In this paper, we propose CPNText-DETR, an innovative method that addresses these challenges. Specifically, we introduce a mass-based dynamic point query modeling strategy, which dynamically initializes contour points based on the mass center of the bounding box, aligning them more closely with the actual text shape. Additionally, we present the Contour Point Normalized (CPN) loss, which ensures uniform weighting for contour point regression across texts of varying scales, thus improving regression balance. Furthermore, we introduce the Parallel Enhanced Factorized Self-Attention (PEFSA) module, which captures intra- and inter-textual relationships, enhancing both performance and inference speed. Extensive experiments demonstrate that CPNText-DETR achieves state-of-the-art performance, with F-measures of 89.6%, 89.5%, 90.5% and 80.1% on the Total-Text, CTW1500, ICDAR2015 and Art datasets, respectively. CPNText-DETR not only surpasses existing methods in detection accuracy but also offers faster inference speed, establishing new benchmarks across widely-used datasets. The code is available at https://github.com/youngnuaa/CPNText-DETR.

最近基于变换的任意形状文本检测方法通过预测文本轮廓点显示出改进的性能。然而，这些方法往往依赖于边界框中心进行文本轮廓点定位，可能不准确地落在文本区域内，并且使用L1损失进行轮廓点回归，导致尺度之间的不平衡。在本文中，我们提出了一种解决这些挑战的创新方法CPNText-DETR。具体来说，我们引入了一种基于质量的动态点查询建模策略，该策略根据边界框的质量中心动态初始化轮廓点，使轮廓点与实际文本形状更接近。此外，我们提出了轮廓点归一化（CPN）损失，它确保了不同尺度文本之间轮廓点回归的均匀加权，从而改善了回归平衡。此外，我们还引入了并行增强因子化自关注（PEFSA）模块，该模块捕获了文本内和文本间的关系，从而提高了性能和推理速度。大量的实验表明，CPNText-DETR达到了最先进的性能，在Total-Text、CTW1500、ICDAR2015和Art数据集上的f值分别为89.6%、89.5%、90.5%和80.1%。CPNText-DETR不仅在检测精度上超越了现有方法，而且提供了更快的推理速度，在广泛使用的数据集上建立了新的基准。代码可在https://github.com/youngnuaa/CPNText-DETR上获得。

{"title":"CPNText-DETR: Dynamic contour initialization and normalized loss for accurate arbitrarily shaped text detection","authors":"Yang Wang , Wanlin Zhou , Chen Lv , Hongjun Xia , Jiakai Zhou","doi":"10.1016/j.cviu.2026.104647","DOIUrl":"10.1016/j.cviu.2026.104647","url":null,"abstract":"<div><div>Recent transformer-based methods for arbitrarily shaped text detection have shown improved performance by predicting text contour points. However, these methods often rely on bounding box centers for text contour points localization, which may not accurately fall within the text area, and use L1 loss for contour point regression, causing imbalances across scales. In this paper, we propose CPNText-DETR, an innovative method that addresses these challenges. Specifically, we introduce a mass-based dynamic point query modeling strategy, which dynamically initializes contour points based on the mass center of the bounding box, aligning them more closely with the actual text shape. Additionally, we present the Contour Point Normalized (CPN) loss, which ensures uniform weighting for contour point regression across texts of varying scales, thus improving regression balance. Furthermore, we introduce the Parallel Enhanced Factorized Self-Attention (PEFSA) module, which captures intra- and inter-textual relationships, enhancing both performance and inference speed. Extensive experiments demonstrate that CPNText-DETR achieves state-of-the-art performance, with F-measures of 89.6%, 89.5%, 90.5% and 80.1% on the Total-Text, CTW1500, ICDAR2015 and Art datasets, respectively. CPNText-DETR not only surpasses existing methods in detection accuracy but also offers faster inference speed, establishing new benchmarks across widely-used datasets. The code is available at <span><span>https://github.com/youngnuaa/CPNText-DETR</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50633,"journal":{"name":"Computer Vision and Image Understanding","volume":"265 ","pages":"Article 104647"},"PeriodicalIF":3.5,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191768","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Robust video anomaly detection via causal feature-guided data augmentation 基于因果特征引导数据增强的鲁棒视频异常检测

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-01-29 DOI: 10.1016/j.cviu.2026.104671

Zhaoyi Li , Chenrui Shi , Che Sun , Yuwei Wu

Data augmentation is a commonly used technique to learn distinguishing features for accurately identifying video anomaly patterns. However, traditional data augmentation methods introduce spurious features-spurious correlations (e.g., lighting conditions) that lack causal links to anomalies, yielding models that excel on specific test sets but fail in real-world scenarios with varying conditions. In this paper, we propose a novel robust video anomaly detection method by designing a causal feature-guided augmentation technique that strategically enhances the learnability of predictive patterns while suppressing spurious correlations. Specifically, we first disentangle causal features ,i.e., directly predictive of labels, from spurious features via a causal generative model. We then perturb causal features to enhance their variability in training data, compelling the model to focus on invariant patterns while ignoring spurious correlations. The augmented samples are reconstructed with the refined causal representations, enhancing the model’s discriminative capability. Furthermore, we introduce a systematic evaluation framework with three increasing difficulty levels to assess robustness: seen/unseen variations, cross-dataset generalization, and cross-domain adaptation. This evaluation examines system stability under varying conditions, closely aligning with real-world surveillance deployment requirements. Experimental results validate both the effectiveness and robustness of our method.

为了准确识别视频异常模式，数据增强是一种常用的特征学习技术。然而，传统的数据增强方法引入了虚假的特征-虚假的相关性（例如，照明条件），缺乏与异常的因果联系，产生的模型在特定的测试集上表现出色，但在具有不同条件的真实场景中失败。在本文中，我们提出了一种新的鲁棒视频异常检测方法，通过设计一种因果特征引导的增强技术，在抑制虚假相关性的同时战略性地增强预测模式的可学习性。具体来说，我们首先解开因果特征，即：通过因果生成模型从虚假特征中直接预测标签。然后，我们扰动因果特征以增强其在训练数据中的可变性，迫使模型专注于不变模式而忽略虚假相关性。利用改进的因果表示重构扩充样本，增强了模型的判别能力。此外，我们引入了一个系统评估框架，该框架具有三个增加难度的级别来评估鲁棒性：可见/未见变化，跨数据集泛化和跨领域适应。该评估检查了系统在不同条件下的稳定性，与现实世界的监视部署需求密切一致。实验结果验证了该方法的有效性和鲁棒性。

{"title":"Robust video anomaly detection via causal feature-guided data augmentation","authors":"Zhaoyi Li , Chenrui Shi , Che Sun , Yuwei Wu","doi":"10.1016/j.cviu.2026.104671","DOIUrl":"10.1016/j.cviu.2026.104671","url":null,"abstract":"<div><div>Data augmentation is a commonly used technique to learn distinguishing features for accurately identifying video anomaly patterns. However, traditional data augmentation methods introduce spurious features-spurious correlations (<em>e.g.</em>, lighting conditions) that lack causal links to anomalies, yielding models that excel on specific test sets but fail in real-world scenarios with varying conditions. In this paper, we propose a novel robust video anomaly detection method by designing a causal feature-guided augmentation technique that strategically enhances the learnability of predictive patterns while suppressing spurious correlations. Specifically, we first disentangle causal features ,<em>i.e.</em>, directly predictive of labels, from spurious features via a causal generative model. We then perturb causal features to enhance their variability in training data, compelling the model to focus on invariant patterns while ignoring spurious correlations. The augmented samples are reconstructed with the refined causal representations, enhancing the model’s discriminative capability. Furthermore, we introduce a systematic evaluation framework with three increasing difficulty levels to assess robustness: seen/unseen variations, cross-dataset generalization, and cross-domain adaptation. This evaluation examines system stability under varying conditions, closely aligning with real-world surveillance deployment requirements. Experimental results validate both the effectiveness and robustness of our method.</div></div>","PeriodicalId":50633,"journal":{"name":"Computer Vision and Image Understanding","volume":"265 ","pages":"Article 104671"},"PeriodicalIF":3.5,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081622","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

CSN: A compact semantic segmentation network for visual scene perception in assistive navigation 一种用于辅助导航视觉场景感知的紧凑语义分割网络

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-01-22 DOI: 10.1016/j.cviu.2026.104665

Yunjia Lei , Son Lam Phung , Yang Di , Abdesselam Bouzerdoum

Accuracy and efficiency are essential in assistive navigation algorithms to ensure accessibility and reliability for visually impaired individuals. However, existing deep learning models often require substantial computational resources to achieve high accuracy, making them impractical for deployment on mobile devices. To address this problem, we introduce CSN, a compact semantic segmentation network designed for assistive navigation, optimizing performance in resource-constrained environments. With CSN, we introduce two innovative modules, the cascaded atrous multi-scale enhancement (CAME) layer and the dual-path residual bottleneck (DPRB) block. The CAME layer efficiently enhances multi-scale representation through feature resampling, while the DPRB block improves feature refinement with minimal computational cost. These modules enable CSN to achieve robust and reliable segmentation across diverse and complex pedestrian environments. The proposed approach achieves the best result on the challenging TrueSight dataset, demonstrating superior prediction accuracy and computational efficiency compared to state-of-the-art lightweight models. CSN achieves a mean intersection over union of 60.99%, while maintaining a low computational cost of 83.46 giga floating-point operations, a compact model size of 8.41 million parameters, and a real-time inference speed of 52.59 frames per second.

准确性和效率是辅助导航算法的关键，以确保视障人士的可访问性和可靠性。然而，现有的深度学习模型通常需要大量的计算资源来实现高精度，这使得它们在移动设备上的部署不切实际。为了解决这个问题，我们引入了CSN，一种紧凑的语义分割网络，用于辅助导航，优化资源受限环境下的性能。在CSN中，我们引入了两个创新模块，级联的多尺度增强（CAME）层和双路剩余瓶颈（DPRB）块。CAME层通过特征重采样有效地增强了多尺度表示，而DPRB块以最小的计算成本提高了特征细化。这些模块使CSN能够在各种复杂的行人环境中实现稳健可靠的分割。该方法在具有挑战性的TrueSight数据集上取得了最佳结果，与最先进的轻量级模型相比，展示了更高的预测精度和计算效率。CSN实现了60.99%的平均交并，同时保持了83.46千兆浮点运算的低计算成本，841万个参数的紧凑模型规模，52.59帧/秒的实时推理速度。

{"title":"CSN: A compact semantic segmentation network for visual scene perception in assistive navigation","authors":"Yunjia Lei , Son Lam Phung , Yang Di , Abdesselam Bouzerdoum","doi":"10.1016/j.cviu.2026.104665","DOIUrl":"10.1016/j.cviu.2026.104665","url":null,"abstract":"<div><div>Accuracy and efficiency are essential in assistive navigation algorithms to ensure accessibility and reliability for visually impaired individuals. However, existing deep learning models often require substantial computational resources to achieve high accuracy, making them impractical for deployment on mobile devices. To address this problem, we introduce CSN, a compact semantic segmentation network designed for assistive navigation, optimizing performance in resource-constrained environments. With CSN, we introduce two innovative modules, the cascaded atrous multi-scale enhancement (CAME) layer and the dual-path residual bottleneck (DPRB) block. The CAME layer efficiently enhances multi-scale representation through feature resampling, while the DPRB block improves feature refinement with minimal computational cost. These modules enable CSN to achieve robust and reliable segmentation across diverse and complex pedestrian environments. The proposed approach achieves the best result on the challenging TrueSight dataset, demonstrating superior prediction accuracy and computational efficiency compared to state-of-the-art lightweight models. CSN achieves a mean intersection over union of 60.99%, while maintaining a low computational cost of 83.46 giga floating-point operations, a compact model size of 8.41 million parameters, and a real-time inference speed of 52.59 frames per second.</div></div>","PeriodicalId":50633,"journal":{"name":"Computer Vision and Image Understanding","volume":"264 ","pages":"Article 104665"},"PeriodicalIF":3.5,"publicationDate":"2026-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146038532","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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

Biometric technology roadmapping for personalized augmentative and alternative communication 个性化、增强和替代通信的生物识别技术路线图

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-01-22 DOI: 10.1016/j.cviu.2026.104664

Svetlana Yanushkevich , Eva Berepiki , Philip Ciunkiewicz , Vlad Shmerko , Gregor Wolbring , Richard Guest

The purpose of this paper is to provide a biometric based technology roadmap to advance personalized Augmentative and Alternative Communication (AAC) systems for individuals with disabilities. The proposed technology roadmap introduces two core components: an AAC biometric register and interoperable technological modules. The biometric register provides a structured framework for capturing and transforming physiological and behavioral traits to enable adaptive and context-aware communication. The interoperable module design supports reconfigurable AAC architectures, ensuring compatibility across diverse interfaces and devices. Together, these mechanisms establish a foundation for automated and scalable personalization in AAC technologies. The study demonstrates that the proposed methodology for technology roadmapping effectively connects established research with emerging computational practices. This is confirmed by the results of the case study such as sign language recognition.

本文的目的是提供一个基于生物识别的技术路线图，以推进残疾人个性化的辅助和替代通信（AAC）系统。提出的技术路线图引入了两个核心组件：AAC生物识别寄存器和可互操作的技术模块。生物识别寄存器提供了一个结构化的框架，用于捕获和转换生理和行为特征，以实现自适应和上下文感知的通信。可互操作的模块设计支持可重构的AAC架构，确保跨不同接口和设备的兼容性。总之，这些机制为AAC技术中的自动化和可伸缩个性化奠定了基础。研究表明，提出的技术路线图方法有效地将已建立的研究与新兴的计算实践联系起来。手语识别等案例研究的结果也证实了这一点。

引用次数: 0

HGAvatar: One-shot high-quality 3D Gaussian head avatar HGAvatar：一次拍摄的高质量3D高斯头像

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Computer Vision and Image Understanding

Pub Date : 2026-01-21 DOI: 10.1016/j.cviu.2026.104660

Xueping Wang, Jun Xu, Feihu Yan, Guangzhe Zhao

3D Gaussian Splatting (3DGS) has sparked growing interest among researchers in 3D head avatars, which is an important topic in the fields of digital human and artificial intelligence. Due to the limitations of monocular videos and the complexity of facial textures and motions, creating high-fidelity 3D head avatars still poses significant challenges. In this paper, we propose a High-Quality 3D Gaussian Head Avatar (HGAvatar) that achieves modeling head avatars with realistic textures and accurate poses within the limitations of monocular video. Specifically, we design a Detail Optimization Module (DOM) to enhance the capability in capturing intricate facial texture details. Additionally, we introduce a Geometric Deformation Module (GDM), which decouples head pose from facial expression and utilizes them as guiding factors to further adjust the properties of the Gaussian ellipsoid, allowing the model to learn more accurate fine details of facial motions. Both qualitative and quantitative experiments demonstrate that the proposed method can better preserve the fine facial texture details and accurately reproduce subtle differences in facial dynamics compared to the state-of-the-art methods.

三维高斯溅射（3DGS）是数字人类和人工智能领域的一个重要课题，引起了人们对三维头像研究的兴趣。由于单目视频的局限性和面部纹理和运动的复杂性，创建高保真3D头部头像仍然面临重大挑战。在本文中，我们提出了一种高质量的3D高斯头部头像（HGAvatar），在单目视频的限制下实现了具有逼真纹理和准确姿态的头部头像建模。具体来说，我们设计了一个细节优化模块（DOM）来增强对复杂面部纹理细节的捕获能力。此外，我们还引入了一个几何变形模块（GDM），该模块将头部姿态与面部表情解耦，并利用它们作为指导因素进一步调整高斯椭球的属性，使模型能够更准确地学习面部运动的精细细节。定性和定量实验均表明，与现有方法相比，该方法能更好地保留面部纹理细节，准确再现面部动态的细微差异。

{"title":"HGAvatar: One-shot high-quality 3D Gaussian head avatar","authors":"Xueping Wang, Jun Xu, Feihu Yan, Guangzhe Zhao","doi":"10.1016/j.cviu.2026.104660","DOIUrl":"10.1016/j.cviu.2026.104660","url":null,"abstract":"<div><div>3D Gaussian Splatting (3DGS) has sparked growing interest among researchers in 3D head avatars, which is an important topic in the fields of digital human and artificial intelligence. Due to the limitations of monocular videos and the complexity of facial textures and motions, creating high-fidelity 3D head avatars still poses significant challenges. In this paper, we propose a High-Quality 3D Gaussian Head Avatar (HGAvatar) that achieves modeling head avatars with realistic textures and accurate poses within the limitations of monocular video. Specifically, we design a Detail Optimization Module (DOM) to enhance the capability in capturing intricate facial texture details. Additionally, we introduce a Geometric Deformation Module (GDM), which decouples head pose from facial expression and utilizes them as guiding factors to further adjust the properties of the Gaussian ellipsoid, allowing the model to learn more accurate fine details of facial motions. Both qualitative and quantitative experiments demonstrate that the proposed method can better preserve the fine facial texture details and accurately reproduce subtle differences in facial dynamics compared to the state-of-the-art methods.</div></div>","PeriodicalId":50633,"journal":{"name":"Computer Vision and Image Understanding","volume":"265 ","pages":"Article 104660"},"PeriodicalIF":3.5,"publicationDate":"2026-01-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081623","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"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