检索结果-内蒙古大学图书馆

IEEE/CVF conference on Computer Vision and Pattern Recognition (CVPR)

作者： Li, Sheng Tan, Mingxing Pang, Ruoming Li, Andrew Cheng, Liqun Le, Quoc, V Jouppi, Norman P. Google Mountain View CA 94043 USA

ISBN: (纸本)9781665445092

Neural Architecture Search (NAS), together with model scaling, has shown remarkable progress in designing high accuracy and fast convolutional architecture families. However, as neither NAS nor model scaling considers sufficient hardware architecture details, they do not take full advantage of the emerging datacenter (DC) accelerators. In this paper, we search for fast and accurate CNN model families for efficient inference on DC accelerators. We first analyze DC accelerators and find that existing CNNs suffer from insufficient operational intensity, parallelism, and execution efficiency and exhibit FLOPs-latency nonproportionality. These insights let us create a DC-accelerator-optimized search space, with space-to-depth, space-to-batch, hybrid fused convolution structures with vanilla and depthwise convolutions, and block-wise activation functions. We further propose a latency-aware compound scaling (LACS), the first multi-objective compound scaling method optimizing both accuracy and latency. Our LACS discovers that network depth should grow much faster than image size and network width, which is quite different from the observations from previous compound scaling. With the new search space and LACS, our search and scaling on datacenter accelerators results in a new model series named EfficientNet-X. EfficientNet-X is up to more than 2X faster than Efficient-Net (a model series with state-of-the-art trade-off on FLOPs and accuracy) on TPUv3 and GPUv100, with comparable accuracy. EfficientNet-X is also up to 7X faster than recent RegNet and ResNeSt on TPUv3 and GPUv100.

关键词： Convolutional codes Computer vision Convolution Computational modeling Search methods parallel processing Hardware

来源：评论

学校读者我要写书评

暂无评论

Fast parallel bounding volume hierarchy construction

Fast parallel bounding volume hierarchy construction

引用

2020 Asia-Pacific conference on image processing, Electronics and Computers, IPEC 2020

作者： Guo, Xiaozi Zhang, Juan Zhou, Mingquan Shaanxi Normal University School of Computer Science Xi'an China Shaanxi Normal University School of Journalism and Communication Xi'an China

ISBN: (纸本)9781728160665

Our algorithm optimized the parallel structure and speeded up the parallel construction based on the locally-ordered clustering to build the bounding volume hierarchy (BVH). We use the bottom-up approach to build the BVH, merging nearest clusters and subnearest clusters in each iteration. Our algorithm was verified in CUDA and evaluated in GPU ray tracing scenarios. Compared with other methods, our method is more efficient in building rate and effectively generates a more balanced BVH under complex scenarios. © 2020 IEEE.

关键词： Ray tracing

来源：评论

学校读者我要写书评

暂无评论

Learning Fair Representations through Uniformly distributed Sensitive Attributes

Learning Fair Representations through Uniformly Distributed ...

引用

Secure and Trustworthy Machine Learning (SaTML), IEEE conference on

作者： Patrik Joslin Kenfack Adín Ramírez Rivera Adil Mehmood Khan Manuel Mazzara Machine Learning and Knowledge Representation Lab Innopolis University Innopolis Russia Department of Informatics Digital Signal Processing and Image Analysis (DSB) group University of Oslo Oslo Norway School of Computer Science University of Hull Hull UK Institute of Software Development and Engineering Innopolis University Innopolis Russia

Machine Learning (ML) models trained on biased data can reproduce and even amplify these biases. Since such models are deployed to make decisions that can affect people's lives, ensuring their fairness is critical. One approach to mitigate possible unfairness of ML models is to map the input data into a less-biased new space by means of training the model on fair representations. Several methods based on adversarial learning have been proposed to learn fair representation by fooling an adversary in predicting the sensitive attribute (e.g., gender or race). However, adversarial-based learning can be too difficult to optimize in practice; besides, it penalizes the utility of the representation. Hence, in this research effort we train bias-free representations from the input data by inducing a uniform distribution over the sensitive attributes in the latent space. In particular, we propose a probabilistic framework that learns these representations by enforcing the correct reconstruction of the original data, plus the prediction of the attributes of interest while eliminating the possibility of predicting the sensitive ones. Our method leverages the inability of Deep Neural Networks (DNNs) to generalize when trained on a noisy label space to regularize the latent space. We use a network head that predicts a noisy version of the sensitive attributes in order to increase the uncertainty of their predictions at test time. Our experiments in two datasets demonstrated that the proposed model significantly improves fairness while maintaining the prediction accuracy of downstream tasks.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Underwater LED light MIMO telecentric optical path technology research

Underwater LED light MIMO telecentric optical path technolog...

引用

IEEE International Symposium on parallel and distributed processing with Applications and IEEE International conference on Ubiquitous Computing and Communications (ISPA/IUCC)

作者： Jiangnan Liu Guojin Peng Jiapeng Su Provincial Ministry of Education Key Laboratory of Cognitive Radio and Information Processing Guilin University of Electronic Technology Guilin China

The system capacity can be further improved by building Light Emitting Diode (LED) light Multiple-Input Multiple-Output (MIMO) on the basis of hemispherical array structure, but the MIMO capacity is limited by the degree of sub-channel interference. To reduce the sub-channel interference of hemispherical arrays, this paper proposes a method to control the sub-channel interference of LED optical MIMO based on the telecentric optical path, which can achieve the sub-channel isolation of LED optical MIMO in hemispherical array structure by using the restriction effect of the optical diaphragm on stray light and maintain the stable sub-channel isolation in a large communication range. Compared with the conventional Singular Value Decomposition (SVD)/Space-time Block Coding (STBC)/ optical zoom and other electrical and optical domain LED optical MIMO sub-channel interference control methods, the proposed method has better Bit Error Ratio (BER) performance and lower structural complexity, and improves the robustness of sub-channel correlation to communication distance variation.

关键词： Stimulated emission Interference Stray light Light emitting diodes Ubiquitous computing Adaptive optics Encoding

来源：评论

学校读者我要写书评

暂无评论

End-to-End Medical image Denoising via Cycle-consistent Generative Adversarial Network

End-to-End Medical Image Denoising via Cycle-consistent Gene...

引用

Information Science, parallel and distributed Systems (ISPDS), International conference on

作者： Chenggeng Yan Hu Chen Zhao Yang College of Computer Science Sichuan University Chengdu China National Key Laboratory of Fundamental Science on Synthetic Vision Sichuan University Chengdu China

ISBN: (纸本)9781665417396

Since the radiation of X-ray is detrimental to patients, low-dose computed tomography (CT) has been developed in medical field. Nevertheless, it may degrade the quality of CT images and affect the clinical diagnosis. To address this issue, various advanced low-dose CT denoising methods have been developed to improve the quality of CT images. The majority of these methods require well-matched low-dose CT and normal-dose CT images for the training. These methods have made much progress in low-dose CT denoising field. However, it is not practicable in medical field to acquire well-paired images in medical field which restricts the application of these methods. To deal with this matter, we propose an improved end-to-end CT image denoising without aligned low-dose and normal-dose CT images inspired by Cycle-consistent Generative Adversarial Network. This method can learn the data distributions of low-dose CT and normal-dose CT in order to build a mapping from low-dose CT images to normal-dose CT images without paired CT images. Additionally, the perceptual loss is integrated into the proposed method to enhance the visual quality of denoised images. The experiments on the real clinic CT dataset show our methods can not only suppress noise but also preserve structural ***-Applied computing~Computers in other domains~Computing in government~E-government

关键词： Training Measurement Visualization Computed tomography Noise reduction Generative adversarial networks Modems

来源：评论

学校读者我要写书评

暂无评论

Approximate Nearest-Neighbour Fields via Massively-parallel Propagation-Assisted K-D Trees 8

Approximate Nearest-Neighbour Fields via Massively-Parallel ...

引用

8th IEEE International conference on Big Data (Big Data)

作者： Oancea, Cosmin Eugen Robroek, Ties Gieseke, Fabian Univ Copenhagen Dept Comp Sci DIKU Copenhagen Denmark Univ Munster Munster Germany

ISBN: (纸本)9781728162515

Nearest neighbour fields accurately and intuitively describe the transformation between two images and have been heavily used in computer vision. Generating such fields, however, is not an easy task due to the induced computational complexity, which quickly grows with the sizes of the images. Modern parallel devices such as graphics processing units depict a viable way of reducing the practical run time of such compute-intensive tasks. In this work, we propose a novel parallel implementation for one of the state-of-the-art methods for the computation of nearest neighbour fields, called propagation-assisted k-d trees. The resulting implementation yields valuable computational savings over a corresponding multi-core implementation. Additionally, it is tuned to consume only little additional memory and is, hence, capable of dealing with high-resolution image data, which is vital as image quality standards keep rising.

关键词： nearest neighbour fields propagation-assisted k-d trees high-performance computing computer vision

来源：评论

学校读者我要写书评

暂无评论

FFDNet-Based Channel Estimation for Beamspace mmWave Massive MIMO Systems

FFDNet-Based Channel Estimation for Beamspace mmWave Massive...

引用

IEEE International conference on Big Data and Cloud Computing (BdCloud)

作者： Wei Gao Meihong Yang Wei Zhang Yuhan Zhao Yan Zhou Kai Zhang Shandong Provincial Key Laboratory of Computer Networks Shandong Computer Science Center National Supercomputer Center in Jinan) Qilu University of Technology (Shandong Academy of Sciences) Jinan China

In beamspace millimeter-wave massive multiple input and multiple-output systems, the channel estimation can be regarded as a 2D image, and the image processing method is used to solve the problems in channel estimation. The DnCNN denoising proposed in the past has poor flexibility and low efficiency in actual channel estimation. In order to solve this problem, this paper quotes a fast and flexible denoising convolutional neural network FFDNet combined with AMP to apply to the channel estimation of the beamspace mmWave MIMO system. Simulation result representation, even when the receiver is equipped with fewer RF chains, the FFDAMP proposed in the beamspace mmWave MIMO system is superior to several existing channel estimation algorithms in the corresponding noise level.

关键词： Radio frequency Simulation image processing Noise reduction Millimeter wave technology Channel estimation Receivers

来源：评论

学校读者我要写书评

暂无评论

Hisui: an image and Video processing Framework with Auto-optimizer

Hisui: an Image and Video Processing Framework with Auto-opt...

引用

Asia-Pacific Signal and Information processing Association Annual Summit and conference (APSIPA)

作者： Ippei Okuda Masahiro Takaoka Tomoaki Tsumura Nagoya Institute of Technology Japan

ISBN: (纸本)9781665441629

In order to develop a high-performance image and video processing program, it is necessary to tune the program to a target platform. However, this task not only requires a high level of knowledge of computer architecture and parallel processing, but also requires an enormous amount of development time. It makes the cost of program development a significant problem. To solve this problem, we are developing Hisui that aims to achieve both high productivity and high performance in the development of image and video processing programs. Hisui consists of a highly abstract Hisui language and Hisui compiler that compiles programs written using Hisui language. In this study, we design an automatic optimization function for Hisui and implement it on Hisui compiler. This function can automatically apply multi-threading, vectorization, and cache blocking. In order to design this function, we first investigated these optimization methods and identified the appropriate optimization application criteria and methods. Then, based on the results of the survey, we design an automatic optimization function and implement it on Hisui compiler. As a result of evaluating the usefulness of the improved image and video processing framework, we confirmed that the programs optimized by Hisui compiler are up to 3.0 times faster than those optimized by a combination of Halide and its Auto-scheduler.

关键词： Productivity Program processors Costs image processing Optimization methods Computer architecture Information processing

来源：评论

学校读者我要写书评

暂无评论

Lightweight HI source finding for next generation radio surveys

引用

ASTRONOMY AND COMPUTING 2022年 41卷

作者： Tolley, E. Korber, D. Galan, A. Peel, A. Sargent, M. T. Kneib, J. -P. Courbin, F. Starck, J. -L. Ecole Polytech Fed Lausanne EPFL Lab Astrophys Observ Sauverny CH-1290 Versoix Switzerland Ecole Polytech Fed Lausanne EPFL SCITAS CH-1015 Lausanne Switzerland Int Space Sci Inst Hallerstr 6 CH-3012 Bern Switzerland Univ Sussex Astron Ctr Dept Phys & Astron Brighton BN19QH E Sussex England CEA Saclay Dept Astrophys F-91191 Gif Sur Yvette France Ecole Polytech Fed Lausanne EPFL Lab Astrophys Observ Sauverny CH-1290 Versoix Switzerland

Future deep HI surveys will be essential for understanding the nature of galaxies and the content of the Universe. However, the large volume of these data will require distributed and automated processing techniques. We introduce LiSA, a set of python modules for the denoising, detection and characterization of HI sources in 3D spectral data. LiSA was developed and tested on the Square Kilometer Array Science Data Challenge 2 dataset, and contains modules and pipelines for easy domain decomposition and parallel execution. LiSA contains algorithms for 2D-1D wavelet denoising using the starlet transform and flexible source finding using null-hypothesis testing. These algorithms are lightweight and portable, needing only a few user-defined parameters reflecting the resolution of the data. LiSA also includes two convolutional neural networks developed to analyze data cubes which separate HI sources from artifacts and predict the HI source properties. All of these components are designed to be as modular as possible, allowing users to mix and match different components to create their ideal pipeline. We demonstrate the performance of the different components of LiSA on the SDC2 dataset, which is able to find 95% of HI sources with SNR > 3 and accurately predict their properties. (C) 2022 The Author(s). Published by Elsevier B.V.

关键词： methods Data analysis Techniques image processing

来源：评论

学校读者我要写书评

暂无评论

Review and Comparative Analysis of parallel Video Encoding Techniques for VVC 43

Review and Comparative Analysis of Parallel Video Encoding T...

引用

conference on Applications of Digital image processing XLIII

作者： Belememis, Panagiotis Panagou, Natalia Loukopoulos, Thanasis Koziri, Maria Univ Thessaly Sch Sci Dept Comp Sci & Telecommun Papasiopoulou 2-4PO 35100 Lamia Greece Univ Thessaly Sch Sci Dept Comp Sci & Biomed Informat Papasiopoulou 2-4PO 35100 Lamia Greece

ISBN: (数字)9781510638273

ISBN: (纸本)9781510638273

In this paper we review and summarize research results concerning video encoding parallelization, with a primary focus on medium and fine grained methods that operate at block or inner block levels. Taxonomies are illustrated wherever applicable with emphasis to scalability issues. Given the reported results, we turn our attention into the problem of allocating resources (processing cores) to parallel tasks performed by the encoder so as to achieve high speedup. We advocate that a parallelization scheme taking advantage of independently coded areas (e.g., tiles), wavefront parallelism within each area and inner block parallelism at the CTU compression level, can achieve significantly higher parallelization degree compared to standalone methods. An algorithm is then proposed that takes resource allocation decisions at all the aforementioned levels. Both the proposed algorithm and standalone representative approaches from the relevant literature are evaluated in terms of scalability using CTU coding times recorded by CU split parallelism in VTM 6.2. Results show that the potential scalability of the proposed scheme surpasses alternatives.

关键词： parallel video encoding VVC thread parallelism CU splitting tiles slices wavefront

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：