检索结果-内蒙古大学图书馆

End-to-end object detection method based on shape correction and feature selection cross-attention

Journal of Electronic Imaging 2025年第2期34卷

作者： He, Lixin Ge, Luqing Cheng, Zhi Wang, Xiaofeng Yao, Guangzhuang Hu, Zhi Hefei University School of Artificial Intelligence and Big Data Collaborative Innovation Center for Computer Vision and Pattern Recognition Hefei China

DEtection TRansformer (DETR) and its variant models degrade the object detection performance due to the inability to provide the object position a priori and the lack of shape deviation supervision between prediction boxes and ground truth boxes, which makes some of the shape differences between prediction boxes and ground truth boxes too large. To address this problem, we propose an end-to-end object detection method (SF-DETR) based on shape correction and feature selection cross-attention. First, a feature selection layer is introduced to provide the model with a priori object position by combining classification confidence and intersection over union score to filter high-quality prediction boxes and constrain the computational range of decoder cross-attention. Second, the shape bias is numericalized and a shape correction loss function is proposed to supervise the shape bias and ensure that the shape bias is effectively corrected. Finally, we conducted experiments on three public datasets, MS COCO, Caltech Pedestrians, and BIT-Vehicle, and the experimental results show that the method in this paper not only provides the object position a priori effectively but also reduces the shape bias significantly, which improves the prediction box quality and object detection accuracy. © 2025 SPIE and IS&T.

关键词： cross-attention DEtection TRansformer end-to-end object detection shape correction

来源：评论

学校读者我要写书评

暂无评论

Interaction Confidence Attention for Human–Object Interaction Detection

引用

International Journal of computer vision 2025年 1-20页

作者： Zhang, Hong-Bo Lin, Wang-Kai Su, Hang Lei, Qing Liu, Jing-Hua Du, Ji-Xiang Department of Computer Science and Technology Huaqiao University Xiamen361021 China Fujian Key Laboratory of Big Data Intelligence and Security Huaqiao University Xiamen361021 China Xiamen Key Laboratory of Computer Vision and Pattern Recognition Huaqiao University Xiamen361021 China Fujian Province University Xiamen361021 China

In human–object interaction (HOI) detection task, ensuring that interactive pairs receive higher attention weights while reducing the weight of non-interaction pairs is imperative for enhancing HOI detection accuracy. Guiding attention learning is also a key aspect of existing transformer-based algorithms. To tackle this challenge, this study proposes a novel approach termed Interaction Confidence Score Learning Attention (ICSLA), which introduces weakening and augmentation operations into the original attention weight calculation and feature extraction processes. In ICSLA, feature learning is coupled with confidence score learning, simultaneously. Leveraging ICSLA, a new and universal decoder is devised, establishing a transformer-based one-stage HOI detection architecture. Experimental results demonstrate the effectiveness of the proposed method in improving HOI detection accuracy, offering valuable insights for further optimization of attention mechanisms. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.

关键词： Augmentation operation Human–object interaction detection Interaction confidence score learning attention Weakening operation

来源：评论

学校读者我要写书评

暂无评论

PCGAUNet: Pixel Correlation and Gaussian Attention Driven Network for Text Segmentation 27th

PCGAUNet: Pixel Correlation and Gaussian Attention Driven Ne...

引用

27th International Conference on pattern Recognition, ICPR 2024

作者： Roy, Ayush Palaiahnakote, Shivakumara Pal, Umapada Antonacopoulos, Apostolos Ramachandra, Raghavendra Computer Vision and Pattern Recognition Indian Statistical Institute Kolkata India School of Science Engineering and Environment University of Salford Manchester United Kingdom Norwegian University of Science and Technology Trondheim Norway

ISBN: (纸本)9783031784460

Text-line segmentation is still considered challenging for complex background scene images. The success of text detection and recognition depends on the success of the text segmentation. This study presents a new method for text segmentation to facilitate reliable detection and recognition. Therefore, we introduce a new model called Pixel Correlation and Gaussian Attention Driven Network (PCGAUNet) for text segmentation. To extract pixel correlation, we modified the MultiResUnet architecture, which leverages pixel-wise correlation to effectively highlight foreground pixels. In addition, the proposed model utilizes the prior spatial statistics of bottleneck features to create a learnable Gaussian distribution, which guides the decoder for accurate text segmentation. Experimental results on three standard scene text segmentation datasets, ICDAR13 FST, Total Text, and COCO-TS, show that the proposed model outperforms existing methods. Furthermore, the results for the underwater dataset UTS-55 show that our model is robust and generic. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Image segmentation

来源：评论

学校读者我要写书评

暂无评论

XLSI: A New Xception and Log Polar Transform Based Approach for Scene Text Script Identification 27th

XLSI: A New Xception and Log Polar Transform Based Approach ...

引用

27th International Conference on pattern Recognition, ICPR 2024

作者： Roy, Ayush Palaiahnakote, Shivakumara Pal, Umapada Antonacopoulos, Apostolos Blumenstein, Michael Computer Vision and Pattern Recognition Indian Statistical Institute Kolkata Kolkata India School of Science Engineering and Environment University of Salford Manchester United Kingdom University of Technology Sydney Sydney Australia

ISBN: (纸本)9783031784941

Script identification of text in natural scene images is challenging due to complex backgrounds, arbitrary orientations, different-sized characters, varying fonts, and multiple styles. Most existing methods are not effective in the presence of the above challenges. This paper introduces a new approach based on the Xception architecture and employing the log-polar transformed original image as an additional input, enabling the extraction of cues that are invariant to rotation, scaling, but are sensitive to script. The rationale behind the proposed work is that the combination of global features with text style features makes a significant difference in discriminating between different scripts. To combine the features extracted by Xception from the input image and log the polar transform of the input image, the proposed method introduces a style-enhanced fusion block. In addition, to further improve the performance of script identification, the proposed approach uses a new receptive channel selective focal attention module. Comparative evaluation results on three benchmark datasets, namely CVSI 2015, SIW-13, and MLe2e show that the proposed method outperforms the state-of-the-art in terms of classification rate. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

A New HourGlass Network for Detecting Text in Shaky and Non-shaky Video Frames 27th

A New HourGlass Network for Detecting Text in Shaky and Non-...

引用

27th International Conference on pattern Recognition, ICPR 2024

作者： Halder, Arnab Palaiahnakote, Shivakumara Pal, Umapada Blumenstein, Michael Gornale, Shivanand S. Computer Vision and Pattern Recognition Unit Indian Statistical Institute Kolkata Kolkata India School of Science Engineering and Environment University of Salford Manchester United Kingdom University of Technology Sydney Sydney Australia Department of Computer Science Rani Channamma University Belagavi India

ISBN: (纸本)9783031784972

Shaky and non-shaky videos are quite common in real-time applications such as surveillance and monitoring vehicles and human movements in protected areas. As a result, text detection in such videos is a formidable challenge due to motion blur, noise, shaky cameras, poor quality and poor visibility. In contrast to existing text detection methods, which focus on text detection in scene images or specific types of images, the present work focuses on text detection in both shaky and non-shaky video frames. Inspired by the impressive performance of the HourGlass network for adverse situations, we explore the HourGlass network for successful text detection in shaky, non-shaky video frames and natural scene images. To improve the performance of the HourGlass network, we employ the Real-Time Model (RTMHead) for predicting text precisely and the Cross Stage Partial Network (CSPNet), which is a neck architecture for robust feature fusion. In addition, the integration of the SiLu activation function with the HourGlass network improves the discriminative power ability. To test the efficacy of the proposed method, we conducted experiments on shaky and non-shaky video frames, as well as ICDAR 2015 video frames. Furthermore, to show the effectiveness of the proposed method, we used Total-Text scene images for experimentation. The results on different datasets and a comparative study with the state-of-the-art models show that the proposed model outperforms the existing methods. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Image enhancement

来源：评论

学校读者我要写书评

暂无评论

Self-Supervised Pretraining for Fine-Grained Plankton Recognition

arXiv

引用

arXiv 2025年

作者： Kareinen, Joona Eerola, Tuomas Kraft, Kaisa Lensu, Lasse Suikkanen, Sanna Kälviäinen, Heikki LUT University Computer Vision and Pattern Recognition Laboratory Lappeenranta Finland Finnish Environment Institute Helsinki Finland

Plankton recognition is an important computer vision problem due to plankton’s essential role in ocean food webs and carbon capture, highlighting the need for species-level monitoring. However, this task is challenging due to its fine-grained nature and dataset shifts caused by different imaging instruments and varying species distributions. As new plankton image datasets are collected at an increasing pace, there is a need for general plankton recognition models that require minimal expert effort for data labeling. In this work, we study large-scale self-supervised pretraining for fine-grained plankton recognition. We first employ masked autoencoding and a large volume of diverse plankton image data to pretrain a general-purpose plankton image encoder. Then we utilize fine-tuning to obtain accurate plankton recognition models for new datasets with a very limited number of labeled training images. Our experiments show that self-supervised pretraining with diverse plankton data clearly increases plankton recognition accuracy compared to standard ImageNet pretraining when the amount of training data is limited. Moreover, the accuracy can be further improved when unlabeled target data is available and utilized during the pretraining. © 2025, CC BY.

关键词： Carbon sequestration

来源：评论

学校读者我要写书评

暂无评论

PulmoNetX: A Hybrid vision Transformer Approach for Multi-scale Spatial Feature Reduction in Pneumonia Classification 27th

PulmoNetX: A Hybrid Vision Transformer Approach for Multi-sc...

引用

27th International Conference on pattern Recognition, ICPR 2024

作者： Lasker, Asifuzzaman Ghosh, Mridul Obaidullah, Sk Md Chakraborty, Chandan Roy, Kaushik Pal, Umapada Department of Computer Science and Engineering Aliah University Kolkata700160 India Department of Computer Science Shyampur Siddheswari Mahavidyalaya Howrah711312 India Department of Computer Science and Engineering National Institute of Technical Teachers’ Training and Research Kolkata700106 India Department of Computer Science West Bengal State University Barasat700126 India Computer Vision and Pattern Recognition Unit Indian Statistical Institute Kolkata India

ISBN: (纸本)9783031781650

An innovative deep learning structure, PulmoNetX, integrates the capabilities of Convolutional Neural Networks (CNNs) and vision Transformers (ViTs) to enhance pneumonia detection in chest X-ray imagery. During preprocessing, images are normalized in size, converted to grayscale, and subjected to contrast amplification to emphasize essential features. PulmoNetX employs a hybrid methodology to capture both the local and global characteristics of images, leading to significant advancements in diagnosing different pneumonia types, such as COVID-19-induced, viral, and bacterial pneumonia. Comparative studies reveal that PulmoNetX surpasses leading vision Transformer models in terms of precision, recall, F1-score, and overall accuracy, highlighting its advanced processing abilities and its promise as an effective diagnostic tool in X-ray lung disease detection. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Generalist Segmentation Algorithm for Photoreceptors Analysis in Adaptive Optics Imaging 27th

Generalist Segmentation Algorithm for Photoreceptors Analys...

引用

27th International Conference on pattern Recognition, ICPR 2024

作者： Kulyabin, Mikhail Sindel, Aline Pedersen, Hilde R. Gilson, Stuart Baraas, Rigmor Maier, Andreas Pattern Recognition Lab Department of Computer Science Friedrich-Alexander-Universität Erlangen-Nürnberg Erlangen Germany National Centre for Optics Vision and Eye Care Faculty of Health and Social Sciences University of South-Eastern Norway Kongsberg Norway

ISBN: (纸本)9783031781032

Analyzing the cone photoreceptor pattern in images obtained from the living human retina using quantitative methods can be crucial for the early detection and management of various eye conditions. Confocal adaptive optics scanning light ophthalmoscope (AOSLO) imaging enables visualization of the cones from reflections of waveguiding cone photoreceptors. While there have been significant improvements in automated algorithms for segmenting cones in confocal AOSLO images, the process of labeling data remains labor-intensive and manual. This paper introduces a method based on deep learning (DL) for detecting and segmenting cones in AOSLO images. The models were trained on a semi-automatically labeled dataset of 20 AOSLO batches of images of 18 participants for 0∘, 1∘, and 2∘ from the foveal center. F1 scores were 0.968, 0.958, and 0.954 for 0∘, 1∘, and 2∘, respectively, which is better than previously reported DL approaches. Our method minimizes the need for labeled data by only necessitating a fraction of labeled cones, which is especially beneficial in the field of ophthalmology, where labeled data can often be limited. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Adaptive optics

来源：评论

学校读者我要写书评

暂无评论

A New Impressive and Expressive Features Based Model for Personality Traits Identification 27th

A New Impressive and Expressive Features Based Model for Per...

引用

27th International Conference on pattern Recognition, ICPR 2024

作者： Biswas, Kunal Palaiahnakote, Shivakumara Pal, Umapada Chanda, Sukalpa Wu, Xiao-Jun Jadavpur University Kolkata India School of Science Engineering and Environment University of Salford Manchester United Kingdom Computer Vision and Pattern Recognition Unit Indian Statistical Institute Kolkata India Østfold University College Halden Norway Jiangnan University Wuxi China

ISBN: (纸本)9783031781858

Personality traits identification is challenging due to unpredictable changes in the foreground and background of images. In this work, we propose a new deep learning model for personality traits image classification by exploiting visual and textual information from dating images. We believe that there is a strong correlation between the features (visual + textual) of dating images and personality traits images. The features which draw attention are defined as impressive features of dating images and features that convey emotions are expressive features of personality traits images. This observation motivated us to combine the features of dating images to improve the performance of personality traits images classification. To affirm the above observation, we propose multiple convolutional layers followed by max pool layers which extract features from dating and personality traits images simultaneously. To integrate the strengths of impressive and expressive features, the proposed work introduces a dual fusion approach, which fuses features and modalities at different levels. The experiments are conducted on different standard datasets of personality traits images to demonstrate the effectiveness of impressive features in terms of personality trait image classification. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

DATR: Domain Agnostic Text Recognizer 27th

DATR: Domain Agnostic Text Recognizer

引用

27th International Conference on pattern Recognition, ICPR 2024

作者： Purkayastha, Kunal Sarkar, Shashwat Palaiahnakote, Shivakumara Pal, Umapada Ghosal, Palash Computer Vision and Pattern Recognition Unit Indian Statistical Institute Kolkata India School of Science Engineering and Environment University of Salford Manchester United Kingdom Department of Information Technology Sikkim Manipal Institute of Technology Sikkim Manipal University Sikkim Gangtok India

ISBN: (纸本)9783031784460

Recognizing text extracted from multiple domains is complex and challenging because complexities vary from one domain to another. Most existing methods focus either on natural scene text or specific text type but not text of multiple domains, namely, scene, underwater, and drone texts. In addition, the state-of-the-art models ignore the vital cues that exist in multiple instances of the text. This paper presents a new method called the Student-Teacher-Assistant (STA) network, which involves dual CLIP models to exploit cues in multiple text instances. The model that uses ResNet50 in its image encoder is called helper CLIP, while the model that uses ViT in its image encoder is called primary CLIP. The proposed work processes both models simultaneously to extract visual and textual features through image and text encoders. Our work uses cosine similarity for the randomly chosen input image to detect instances similar to the input image. The input and similar instances are supplied to primary and helper CLIPs for visual and textual feature extraction. The outputs of dual CLIPs are fused in a different way through the alignment step for recognizing text accurately, irrespective of domains. To demonstrate the proposed model’s significance, experiments are conducted on a set of standard natural scene text datasets (regular and irregular), underwater images, and drone images. The results on three different domains show that the proposed model outperforms the state-of-the-art recognition models. The datasets and code for public use in training and testing shall be made available on GitHub. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Underwater imaging

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：