基于多通道分离整合的多尺度单幅图像去雨算法

由于密度不同的雨对图像造成的遮挡不同,图像去雨一直都是一项极具挑战性的任务。目前,基于深度学习的图像去雨算法已经成为主流。然而,多数深度学习的架构都是通过堆叠卷积层来设计的,执行去雨任务后图像仍存在着大小不一的雨痕,这些方法并不能很好地关注训练中雨图的局部信息和上下文信息。为了解决上述问题,本文设计一种基于多通道分离整合的卷积神经网络用于图像去雨。第一步通过通道分离,再利用卷积层间的层级连接,构成多尺度模块,最终将不同通道的输出进行整合。该模块可以增大感受野,探索特征图之间的空间信息,更好地提取特征。第二步利用渐进网络来反复计算挖掘上下文信息,能够很好关联到全局特征。整体模型易于实施,可以端对端训练。在常用的数据集以及自建的自动驾驶雨天数据集上的大量实验表明,本文方法比现有方法取得了明显的改进。     

基于C#语言的电子文件名批量处理及在归档管理中的应用

针对电子文件向档案数据库归档环节,往往需要大量更改文件名的问题,文章梳理了如何借助C#语言编写相应的桌面应用程序,以实现文件名的批量更改和电子文件的顺利入库,从而打通电子文件向档案数据库转入的“最后一公里”,切实提高工作效率,节省人力资源。     

公证书与书证优先主义之论

＂好记性不如烂笔头＂＂最淡的墨水胜过最强的记忆＂是基于长期社会生活实践的经验总结和理性认知。在民事诉讼中,便意味着作为证据的书证的稳定性、可靠性优于人证。这一认知倾向始终影响和指引着世界各国的诉讼立法,并产生了诉讼对书证的不同程度的倚重和依赖,书证优先主义便是最直接最有力的印证。两大法系在各国民事证据制度方面或多或少均有所体现,尤以法国民法典第1 341条最为典型。我国证据立法及相关司法解释中亦有一定程度表现,特别是公证书作为诉讼证据所反映的＂书证优先主义＂最具特色,只是在司法实践中常常被漠视和误读而产生颇多歧见。对两大法系主要国家或地区的＂书证优先主义＂进行比较分析,并以公证书为主要考察对象,引发对我国构建适度书证优先主义证据制度的若干思考。     

基于VSM的文本相似度计算的研究*

文本相似度的计算作为其他文本信息处理的基础和关键,其计算准确率和效率直接影响其他文本信息处理的结果。提出改进的DF算法和TDIDF算法,一方面利用了DF算法具有线性的时间复杂度,比较适合大规模文本处理的特点,并通过适当增加关键词的方法,弥补了其对个别有用信息错误过滤的不足;另一方面,利用特征项在特征选择阶段的权重对TDIDF方法进行加权处理,在不增加开销的情况下扩大了文档集的规模,还提高了相似度计算的精确度。     

工程设计演示文档动态性表述的探讨

在工程投标、方案论证、学术交流中 ,可利用多媒体演示软件制作各类文档。为了充分表达制作人员的思想 ,达到预期目的 ,可采用丰富多彩的动态性表现手法 ,使演示产生很好的效果 ,令听众和观众欣然接受。为此 ,对工程设计演示文档的动态性表述进行了探讨 ,以便能更充分地发挥多媒体演示软件的作用     

Herding Cats

Abstract

With the rapid growth in the number of electronic resources available via the Internet, a variety of methods have been developed to organize and access these objects. Librarians, scholars, and computing engineers have each applied their own techniques to the process. This article examines the strengths and weaknesses of organizational systems developed by each group. The appropriateness of including remotely accessed electronic resources as part of a library collection is discussed, and the success of traditional cataloging methods and structures for describing remote electronic resources is examined. These traditional cataloging techniques are compared to other methods for describing, organizing, and accessing electronic documents such as TEI headers, browsing lists, and robot generated search indexes. For Metacatalogs of the future to be successful tools for organizing and accessing all information, methods must be developed to utilize all existing organizational tools by layering, exchanging, and translating data within a loosely-coupled organizational system.     

Analyzing settings for social identity management on Social Networking Sites: Classification,current state,and proposed developments

The rising prevalence of Social Networking Sites (SNS) and their usage in multiple contexts poses new privacy challenges and increasingly prompts users to manage their online identity. To address privacy threats stemming from interacting with other users on SNS, effective Social Identity Management (SIdM) is a key requirement. It refers to the deliberate and targeted disclosure of personal attribute values to a subset of one's contacts or other users on the SNS. Protection against other entities such as the site operator itself or advertisers and application programmers is not covered by SIdM, but could be incorporated in further refinement steps. Features and settings to perform SIdM have been proposed and subsequently implemented partly by some SNS. Yet, these are often isolated solutions that lack integration into a reference framework that states the requirements for successfully managing one's identity. In this article, such a reference framework of existing and desired SIdM settings is derived from identity theory, literature analysis, and existing SNS. Based thereupon, we examine the SIdM capabilities of prevalent SNS and highlight possible improvements. Lastly, we reason about developing a metric to objectively compare the capability of SNS in regards to their support for SIdM.     

Hypermedia Support for Software Development: A Retrospective Assessment

Hypermedia technologies have been available for about a decade. Beginning with Douglas Engelbart, researchers have become interested in applying hypermedia concepts and technologies to software development. This paper assesses this symbiosis between hypermedia and software development as described by research contributions; by our count, more than one hundred articles. To date, no collection of, or guide to, this literature has been published. A number of significant contributions are classified, described, and appraised — providing a guide into what has been a rich but perhaps under-reported research area. Hypermedia has made significant contributions to software development in three primary areas: coping with massive amounts of information associated with software development projects; establishing and maintaining linkages between various types of software engineering documents; and enabling development engineers to record, document, and preserve knowledge about development thoughts, processes and rationale. These contributions may help direct the focus of future research towards building on, combining and assessing the resulting contributions. A research agenda is proposed along these lines.