检索结果-内蒙古大学图书馆

Learning with asymmetric kernels: Least squares and feature interpretation

学校读者我要写书评

暂无评论

arXiv 2022年

作者： He, Mingzhen He, Fan Shi, Lei Huang, Xiaolin Suykens, Johan A.K. Institute Of Image Processing And Pattern Recognition Shanghai Jiao Tong University Shanghai200240 China Shanghai Key Laboratory For Contemporary Applied Mathematics Fudan University Shanghai200433 China School Of Mathematical Sciences Fudan University Shanghai200433 China KU Leuven LeuvenB-3001 Belgium

Asymmetric kernels naturally exist in real life, e.g., for conditional probability and directed graphs. However, most of the existing kernel-based learning methods require kernels to be symmetric, which prevents the use of asymmetric kernels. This paper addresses the asymmetric kernel-based learning in the framework of the least squares support vector machine named AsK-LS, resulting in the first classification method that can utilize asymmetric kernels directly. We will show that AsK-LS can learn with asymmetric features, namely source and target features, while the kernel trick remains applicable, i.e., the source and target features exist but are not necessarily known. Besides, the computational burden of AsK-LS is as cheap as dealing with symmetric kernels. Experimental results on the Corel database, directed graphs, and the UCI database will show that, in the case asymmetric information is crucial, the proposed AsK-LS can learn with asymmetric kernels and performs much better than the existing kernel methods that have to do symmetrization to accommodate asymmetric kernels. © 2022, CC BY.

关键词： Directed graphs

MambaMIM: Pre-training Mamba with State Space Token Interpolation and its Application to Medical image Segmentation

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Tang, Fenghe Nian, Bingkun Li, Yingtai Jiang, Zihang Yang, Jie Liu, Wei Zhou, S. Kevin School of Biomedical Engineering Division of Life Sciences and Medicine University of Science and Technology of China Anhui Hefei230026 China Suzhou Institute for Advanced Research University of Science and Technology of China Jiangsu Suzhou215123 China Suzhou Institute for Advanced Research USTC Suzhou China Department of Automation Shanghai Jiao Tong University China Institute of Image Processing and Pattern Recognition Shanghai Jiao Tong University China Institute of Medical Robotics Shanghai Jiao Tong University China State Key Laboratory of Precision and Intelligent Chemistry University of Science and Technology of China Anhui Hefei230026 China

Recently, the state space model Mamba has demonstrated efficient long-sequence modeling capabilities, particularly for addressing long-sequence visual tasks in 3D medical imaging. However, existing generative self-supervised learning methods have not yet fully unleashed Mamba’s potential for handling long-range dependencies because they overlook the inherent causal properties of state space sequences in masked modeling. To address this challenge, we propose a general-purpose pre-training framework called MambaMIM, a masked image modeling method based on a novel TOKen-Interpolation strategy (TOKI) for the selective structure state space sequence, which learns causal relationships of state space within the masked sequence. Further, MambaMIM introduces a bottom-up 3D hybrid masking strategy to maintain a masking consistency across different architectures and can be used on any single or hybrid Mamba architecture to enhance its multi-scale and long-range representation capability. We pre-train MambaMIM on a large-scale dataset of 6.8K CT scans and evaluate its performance across eight public medical segmentation benchmarks. Extensive downstream experiments reveal the feasibility and advancement of using Mamba for medical image pre-training. In particular, when we apply the MambaMIM to a customized architecture that hybridizes MedNeXt and Vision Mamba, we consistently obtain the state-of-the-art segmentation performance. The code is available at: https://***/FengheTan9/MambaMIM. Copyright © 2024, The Authors. All rights reserved.

关键词： image segmentation

A shift-insensitive full reference image quality assessment model based on quadratic sum of gradient magnitude and LOG signals

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Chen, Congmin Mou, Xuanqin The Institute of Image Processing and Pattern Recognition Xi’an Jiaotong University Xi’an710049 China

image quality assessment that aims at estimating the subject quality of images, builds models to evaluate the perceptual quality of the image in different applications. Based on the fact that the human visual system (HVS) is highly sensitive to structural information, the edge information extraction is widely applied in different IQA metrics. According to previous studies, the image gradient magnitude (GM) and the Laplacian of Gaussian (LOG) operator are two efficient structural features in IQA tasks. However, most of the IQA metrics achieve good performance only when the distorted image is totally registered with the reference image, but fail to perform on images with small translations. In this paper, we propose an FR-IQA model with the quadratic sum of the GM and the LOG signals, which obtains good performance in image quality estimation considering shift-insensitive property for not well-registered reference and distortion image pairs. Experimental results show that the proposed model works robustly on three large scale subjective IQA databases which contain a variety of distortion types and levels, and stays in the state-of-the-art FR-IQA models no matter for single distortion type or across whole database. Furthermore, we validated that the proposed metric performs better with shift-insensitive property compared with the CW-SSIM metric that is considered to be shift-insensitive IQA so far. Meanwhile, the proposed model is much simple than the CW-SSIM, which is efficient for *** Codes 94A08 Copyright © 2020, The Authors. All rights reserved.

关键词： Laplace transforms

Multi-object tracking with pre-classified detection 5th

学校读者我要写书评

暂无评论

Multi-object tracking with pre-classified detection

5th International Conference on Robot Intelligence Technology and Applications, RiTA 2017

作者： Ren, Siqi Zhou, Yue He, Liming Institute of Image Processing and Pattern Recognition Shanghai Jiao Tong University Shanghai China

ISBN: (纸本)9783319784519

Tracking-by-detection is a popular tracking framework nowadays. The paradigm determines that detections will bring huge impact on the final tracking result. Based on the idea, to improve the tracking precision, we propose a novel algorithm which will first divide the detections into false ones, high uncertainty and low uncertainty ones. After that, we first penalize the false detection boundingboxes and then we construct low and high uncertainty tree for different types of detections. For low uncertainty detections, we construct their tracking trees once, for high uncertainty detections, we adopt an improved MHT to delay the data association decision till the end of the sliding windows and make the decision with all the information in the sliding window. Experiments demonstrate that our algorithm can achieved competitive results with state-of-the-art trackers on some challenging datasets such as MOTChallenge2016 [11]. © Springer International Publishing AG, part of Springer Nature 2019.

关键词： Object detection

SSIM-based CTU-level joint optimal bit allocation and rate distortion optimization

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Li, Yang Mou, Xuanqin Institute of Image Processing and Pattern Recognition Xi'an Jiaotong University Xi'an710049 China

Structural similarity (SSIM)-based distortion DSSIM is more consistent with human perception than the traditional mean squared error DMSE. To achieve better video quality, many studies on optimal bit allocation (OBA) and rate-distortion optimization (RDO) used DSSIM as the distortion metric. However, many of them failed to optimize OBA and RDO jointly based on SSIM, thus causing a non-optimal R-DSSIM performance. This problem is due to the lack of an accurate R-DSSIM model that can be used uniformly in both OBA and RDO. To solve this problem, we propose a DSSIM-DMSE model first. Based on this model, the complex R-DSSIM cost in RDO can be calculated as simpler RDMSE cost with a new SSIM-related Lagrange multiplier. This not only reduces the computation burden of SSIM-based RDO, but also enables the R-DSSIM model to be uniformly used in OBA and RDO. Moreover, with the new SSIM-related Lagrange multiplier in hand, the joint relationship of R-DSSIM-λSSIM (the negative derivative of R-DSSIM) can be built, based on which the R-DSSIM model parameters can be calculated accurately. With accurate and unified R-DSSIM model, SSIM-based OBA and SSIM-based RDO are unified together in our scheme, called SOSR. Compared with the HEVC reference encoder HM16.20, SOSR saves 4%, 10%, and 14% bitrate under the same SSIM in all-intra, hierarchical and non-hierarchical low-delay-B configurations, which is superior to other state-of-the-art schemes. An improved version of this manuscript has been accepted by IEEE Transactions on Broadcasting DOI 10.1109/TBC.2021.3068871. The project page is located at http://***/web/xqmou/sosr. Copyright © 2020, The Authors. All rights reserved.

关键词： Signal distortion

An Self-supervised Learning & Self-attention Based Method for Defects Classification on PCB Surface images

学校读者我要写书评

暂无评论

An Self-supervised Learning & Self-attention Based Method fo...

International Conference on Electronics, Communications and Information Technology (CECIT)

作者： Lijie Zhou Xufeng Ling Shan Zhu Zheng Sun Jie Yang AI School Tianhua College Shanghai Normal University Shanghai China ASIC Motor Corporation Limited China Institute of Image Processing and Pattern Recognition Shanghai Jiao tong University China

ISBN: (纸本)9781665437585

Automatic detection of PCB defects become a difficult work in electronics industry, with the rapid development of Integrated Circuit. A good detection method can effectively improve production efficiency and reduce the costs of quality inspection. The traditional image processing methods have the backwards of complicated system structure, difficult image segmentation, and low detection accuracy, etc. In order to solve the above problems, this paper proposes a method of extracting features by self-supervised learning with supervised learning to complete the classification task. This method uses a deep neural network model based on self-supervised learning technology, uses multi-head self-attention mechanism; automatically recognize single or multiple defect areas in PCB, to extract defects features of PCB, and uses full connection classifier to classify features. The open-source (DeepPCB) is selected as the dataset, including 6 defect categories and non-defect categories. In the training process, data enhancement technology is used to increase the training samples, train the model and make it converge, and realize low error recognition on the verification dataset. The experimental results show that the features extracted by the self-supervised learning method are richer than those extracted by the supervised learning. The composite model formed by combining the self-attention mechanism has high accuracy and good robustness, reaches the classification accuracy of 94.33%, and has high accuracy and good generalization ability.

关键词： Training image segmentation Electronics industry Supervised learning Quality control Feature extraction Robustness

End-To-End Graph-Based Deep Semi-Supervised Learning with Extended Graph Laplacian

学校读者我要写书评

暂无评论

End-To-End Graph-Based Deep Semi-Supervised Learning with Ex...

Chinese Automation Congress (CAC)

作者： Zihao Wang Enmei Tu Meng Zhou Jie Yang Institute of Image Processing and Pattern Recognition Shanghai Jiao Tong University Shanghai China

ISBN: (数字)9781728176871

ISBN: (纸本)9781728176888

The process of labeling samples costs time and resources but unlabeled samples are easier to obtain. Recently, graph-based deep semi-supervised learning (GDSSL) training a deep network using a small number of labeled samples and the abundant unlabeled samples has been demonstrated to be promising on image classification tasks. These methods construct a graph to represent the structure of the input data (or hidden features). The successes of these GDSSL algorithms depend upon the structure of the similarity graph. However, existing GDSSL approaches construct the graph using predefined rules (such as knn graph) or fixed similarity measures (such as Gaussian kernel), which may limit the potential of GDSSL. In this paper, we move further in this direction to propose a novel end-to-end GDSSL approach which fully optimizes the whole graph without such limitations. To this end, we concatenate two neural networks (feature network and similarity network) together to learn the categorical label and semantic similarity, respectively, and train the networks with a new regularization term, the extended graph Laplacian, to minimize a unified objective function. Extensive experiments on several benchmark datasets demonstrate that our approach could outperform existing approaches on image classification. Furthermore, as a side-product, the similarity network could give faithful semantic similarity measure of samples, which is not possessed by other GDSSL approaches.

关键词： Semantics Training Laplace equations Semisupervised learning Task analysis Neural networks Linear programming

Colonoscopic image Synthesis For Polyp Detector Enhancement Via Gan And Adversarial Training

学校读者我要写书评

暂无评论

Colonoscopic Image Synthesis For Polyp Detector Enhancement ...

IEEE International Symposium on Biomedical Imaging

作者： Fan He Sizhe Chen Shuaiyi Li Lu Zhou Haiqin Zhang Haixia Peng Xiaolin Huang Institute of Image Processing and Pattern Recognition Shanghai Jiao Tong University Digestive Endoscopy Center Tongren Hospital Shanghai Jiao Tong University School of Medicine Institute of Medical Robotics Shanghai Jiao Tong University

Computer-aided polyp detection system powered by deep neural networks has achieved high performance but also suffers from data insufficiency. To address this problem, recent researches focus on synthesizing new colonoscopic images by Generative Adversarial Network (GAN). However, the synthesized images follow the same distribution as that of the training dataset, which limits the performance of the detectors re-trained on it. Recent studies show that adversarial examples can expand the data distribution and thus adversarial training can effectively improve the robustness of deep neural networks. Inspired by these two factors, this paper proposes a data augmentation framework to directly produce false negative colonoscopic images via GAN and the adversarial attack. The synthesized polyps are natural and experiments on three popular detectors show that compared with using GAN alone, producing false negative images by the adversarial attack can further improve the performance of the re-trained detectors.

关键词： Training image synthesis Neural networks Deep architecture Detectors Generative adversarial networks Robustness

Hierarchical correlation stereo matching network 2nd

学校读者我要写书评

暂无评论

Hierarchical correlation stereo matching network

2nd Chinese Conference on pattern recognition and Computer Vision, PRCV 2019

作者： Chen, Xuliang Zhou, Yue Institute of Image Processing and Pattern Recognition Shanghai Jiao Tong University Shanghai China

ISBN: (纸本)9783030317256

Recently, stereo matching from a pair of rectified images has been cast as a supervised learning task using the powerful representation of convolutional neural networks. However, existing methods only utilize last feature maps output from Siamese Networks to compute similarity measurement, which are lack of multi-levels similarity information to construct an informative cost volume. To solve this problem, we propose a hierarchical correlation operation to compute similarity of stereo pairs at multiple levels. In addition, to yield accurate disparity in illposed region, we propose a stacked hourglass feature network with dense connections to effectively incorporate context information. Then, hybrid matching cost volume is built leveraging hierarchical correlation features and concatenation features of left and right. 3D CNN encoder-decoder architecture is utilized to regularize the cost volume and regress disparity. Experiments demonstrate that our network achieves competitive performance with state-of-the-art methods on Scene Flow, KITTI 2012, and KITTI 2015 datasets. © Springer Nature Switzerland AG 2019.

关键词： Neural networks