基于深度感知特征提取的室内场景理解

从深度图RGB-D域中联合学习RGB图像特征与3D几何信息有利于室内场景语义分割,然而传统分割方法通常需要精确的深度图作为输入,严重限制了其应用范围。提出一种新的室内场景理解网络框架,建立基于语义特征与深度特征提取网络的联合学习网络模型提取深度感知特征,通过几何信息指导的深度特征传输模块与金字塔特征融合模块将学习到的深度特征、多尺度空间信息与语义特征相结合,生成具有更强表达能力的特征表示,实现更准确的室内场景语义分割。实验结果表明,联合学习网络模型在NYU-Dv2与SUN RGBD数据集上分别取得了69.5%与68.4%的平均分割准确度,相比传统分割方法具有更好的室内场景语义分割性能及更强的适用性。     

跨数据集评估的高光谱图像分类

目的 随着高光谱成像技术的飞速发展,高光谱数据的应用越来越广泛,各场景高光谱图像的应用对高精度详细标注的需求也越来越旺盛。现有高光谱分类模型的发展大多集中于有监督学习,大多数方法都在单个高光谱数据立方中进行训练和评估。由于不同高光谱数据采集场景不同且地物类别不一致,已训练好的模型并不能直接迁移至新的数据集得到可靠标注,这也限制了高光谱图像分类模型的进一步发展。本文提出跨数据集对高光谱分类模型进行训练和评估的模式。方法 受零样本学习的启发,本文引入高光谱类别标签的语义信息,拟通过将不同数据集的原始数据及标签信息分别映射至同一特征空间以建立已知类别和未知类别的关联,再通过将训练数据集的两部分特征映射至统一的嵌入空间学习高光谱图像视觉特征和类别标签语义特征的对应关系,即可将该对应关系应用于测试数据集进行标签推理。结果 实验在一对同传感器采集的数据集上完成,比较分析了语义—视觉特征映射和视觉—语义特征映射方向,对比了5种基于零样本学习的特征映射方法,在高光谱图像分类任务中实现了对分类模型在不同数据集上的训练和评估。结论 实验结果表明,本文提出的基于零样本学习的高光谱分类模型可以实现跨数据集对分类模型进行训练和评估,在高光谱图像分类任务中具有一定的发展潜力。     

为SOFL Specification自动生成消息序列(MSC)

形式化方法以其精确的描述能力,可验证性,和自动化潜力而得到了人们的关注。特别是在一些对安全性要求非常高的领域,已经有了很多成功的案例。但他始终没有在业界得到推广。原因之一就是基于数学符号的形式化语法十分晦涩难懂,这给开发人员之间,开发人员和客户之间的交流造成了巨大的障碍。我们认为要解决这一问题,就需要为形式化规范提供更直观,更为大家所熟悉的行为表现方式。文章介绍如何自动生成MSC(messagesequencechart)来描述SOFL(StructuredObjectorientedFormalLanguage)[12]specification的行为。通过用MSC来动态模拟规范的行为的执行过程,可以很好地帮助人们理解规范内容,这对于规范校验(specificationvalidation)有重要的意义。此外,我们为支持这一MSC自动生成过程编写的一款工具原形,也会在文中作简要介绍。     

Generating semantically valid test inputs using constrained input grammars

ContextGenerating test cases based on software input interface is a black-box testing technique that can be made more effective by using structured input models such as input grammars. Automatically generating grammar-based test inputs may lead to structurally valid but semantically invalid inputs that may be rejected in early semantic error checking phases of a system under test.ObjectiveThis paper aims to introduce a method for specifying a grammar-based input model with the model’s semantic constraints to be used in the generation of positive test inputs. It is also important that the method can generate effective test suites based on appropriate grammar-based coverage criteria.MethodFormal specification of both input structure and input semantics provides the opportunity to use model instantiation techniques to create model instances that satisfy all specified constraints. The input interface of a subject system can be specified using a high-level specification scheme such as attribute grammars, and a transformation function from this scheme to an instantiable formal modeling language can generate the desired model instances.ResultsWe propose a declarative grammar-based input specification method that is based on a variation of attribute grammars and allows the user to specify input constraints in addition to input structure. The model can be instantiated automatically to generate structurally and semantically valid test inputs. The proposed method has the capability to specify test requirements and coverage criteria and use them to generate valid test suites that satisfy test coverage criteria requirements.ConclusionThe work presented in this paper provides a black-box test generation method for grammar-based software inputs that can automatically generate criteria-covering test suites.     

语义特征建模系统中操作局部化的研究

为了提高语义特征建模系统约束求解的效率及系统的整体性能,提出了一种新的操作局部化方法。该方法使用特征语义表示法来描述产品模型中特征的各种信息,改进了细胞元模型对特征元素的管理机制,通过特征的“语义面”将局部特征从模型中有效地分离出来,通过创建临时特征的方法来实现模型的局部约束求解。该策略不仅可以完全满足在复杂模型中直接操作的需要,还可以大大提高系统的性能。实验表明该算法具有更强的适应性和实用性。     

融合语义—表观特征的无监督前景分割

目的 前景分割是图像理解领域中的重要任务,在无监督条件下,由于不同图像、不同实例往往具有多变的表达形式,这使得基于固定规则、单一类型特征的方法很难保证稳定的分割性能。针对这一问题,本文提出了一种基于语义-表观特征融合的无监督前景分割方法（semantic apparent feature fusion,SAFF）。方法 基于语义特征能够对前景物体关键区域产生精准的响应,但往往产生的前景分割结果只关注于关键区域,缺乏物体的完整表达;而以显著性、边缘为代表的表观特征则提供了更丰富的细节表达信息,但基于表观规则无法应对不同的实例和图像成像模式。为了融合表观特征和语义特征优势,研究建立了融合语义、表观信息的一元区域特征和二元上下文特征编码的方法,实现了对两种特征表达的全面描述。接着,设计了一种图内自适应参数学习的方法,用于计算最适合的特征权重,并生成前景置信分数图。进一步地,使用分割网络来学习不同实例间前景的共性特征。结果 通过融合语义和表观特征并采用图像间共性语义学习的方法,本文方法在PASCAL VOC（pattern analysis,statistical modelling and computational learning visual object classes）2012训练集和验证集上取得了显著超过类别激活映射（class activation mapping,CAM）和判别性区域特征融合方法（discriminative regional feature integration,DRFI）的前景分割性能,在F测度指标上分别提升了3.5%和3.4%。结论 本文方法可以将任意一种语义特征和表观特征前景计算模块作为基础单元,实现对两种策略的融合优化,取得了更优的前景分割性能。     

协同语义特征建模技术研究

为了提高协同语义建模系统的性能,提出了一种新的协同设计方法,亦即使用特征语义表示法和细胞元模型来表示数据和管理模型中的各种数据和信息,然后通过语义匹配的方法来确定子模型间的逻辑和装配关系,通过语义区域划分、"最值空间"等技术来检测和消解协同设计中发生的操作冲突,最后通过创建临时模型来降低全局约束求解的复杂度。这种方法不仅能够完全实现协同语义特征建模的需要,还可以大大提高协同设计的效率。实验表明,该算法具有更强的适应性和实用性。     

Interactive quadrangulation with Reeb atlases and connectivity textures

Creating high-quality quad meshes from triangulated surfaces is a highly nontrivial task that necessitates consideration of various application specific metrics of quality. In our work, we follow the premise that automatic reconstruction techniques may not generate outputs meeting all the subjective quality expectations of the user. Instead, we put the user at the center of the process by providing a flexible, interactive approach to quadrangulation design. By combining scalar field topology and combinatorial connectivity techniques, we present a new framework, following a coarse to fine design philosophy, which allows for explicit control of the subjective quality criteria on the output quad mesh, at interactive rates. Our quadrangulation framework uses the new notion of Reeb atlas editing, to define with a small amount of interactions a coarse quadrangulation of the model, capturing the main features of the shape, with user prescribed extraordinary vertices and alignment. Fine grain tuning is easily achieved with the notion of connectivity texturing, which allows for additional extraordinary vertices specification and explicit feature alignment, to capture the high-frequency geometries. Experiments demonstrate the interactivity and flexibility of our approach, as well as its ability to generate quad meshes of arbitrary resolution with high-quality statistics, while meeting the user’s own subjective requirements.     

A Virtual World Grammar for automatic generation of virtual worlds

Hybrid systems such as those that combine 3D virtual worlds and organization based multiagent systems add new visual and communication features for multiuser applications. The design of such hybrid and dynamic systems is a challenging task. In this paper, we propose a system that can automatically generate a 3D virtual world (VW) from an organization based multiagent system (MAS) specification that establishes the activities participants can engage on. Both shape grammar and virtual world paradigms inspired us to propose a Virtual World Grammar (VWG) to support the generation process of a virtual world design. A VWG includes semantic information about both MAS specification and shape grammar elements. This information, along with heuristics and validations, guides the VW generation producing functional designs. To support the definition and execution of a Virtual World Grammar, we have developed a so named Virtual World Builder Toolkit (VWBT). We illustrate this process by generating a 3D visualization of a virtual institution from its specification.     

语义特征造型中特征修改优先准则的研究

特征造型技术是新一代CAD系统的关键技术,是产品模型技术的核心。该文讨论了以细胞元模型为基础的语义特征造型技术。在语义特征造型中用特征依赖图（FDG）的数据模型来保存和维护设计中的各种特征信息及其之间的关系,在此基础上分析了依据特征依赖图中特征间的关系动态修改特征优先级的方法,确保了模型修改的有效性。最后给出了一个实例。     

A hybrid probabilistic framework for content-based image retrieval with feature weighting

In this paper, we propose a probabilistic framework for efficient retrieval and indexing of image collections. This framework uncovers the hierarchical structure underlying the collection from image features based on a hybrid model that combines both generative and discriminative learning. We adopt the generalized Dirichlet mixture and maximum likelihood for the generative learning in order to estimate accurately the statistical model of the data. Then, the resulting model is refined by a new discriminative likelihood that enhances the power of relevant features. Consequently, this new model is suitable for modeling high-dimensional data described by both semantic and low-level (visual) features. The semantic features are defined according to a known ontology while visual features represent the visual appearance such as color, shape, and texture. For validation purposes, we propose a new visual feature which has nice invariance properties to image transformations. Experiments on the Microsoft's collection (MSRCID) show clearly the merits of our approach in both retrieval and indexing.     

一种面向服务的领域特征模型

针对面向服务的领域需求表达与规约化组织,提出一种扩展的服务特征概念与服务特征模型,以及基于OWL本体描述语言的服务特征本体元模型。扩展的服务特征概念可更为完整地表征面向服务领域的服务需求语义,而扩展的服务特征模型则为面向服务领域需求提供了一种易于扩展的规约化组织模型。模型通过特征关系映射,既克服了传统特征模型的特征组织形式的局限性,也丰富了服务特征间的语义关系,并可更好地支持基于特征的领域需求配置管理与定制。最后,通过给出的一个原型辅助建模工具实现,验证了模型的适用性。     

语义特征建模系统中布尔算法的研究

为了提高语义特征建模系统中布尔操作的运行效率,提出了一种基于语义表示法的布尔操作算法.该算法用语义表示法表示特征模型,用细胞元模型组织和管理特征元素,用语义面替代几何面来提高特征的交互检测效率,通过细胞分裂和语义面分解来生成新实体.该算法不仅可以快速准确地生成布尔实体,还可以避免几何面的丢失及“孔洞”等错误的发生.实验证明,该算法具有广泛的使用前景和实用价值.     

面向智能计算框架的即时缺陷预测

作为人工智能工程化的实现工具,智能计算框架已在近年来被广泛应用,其可靠性对于人工智能的有效实现至关重要.然而,智能计算框架的可靠性保障具有挑战性,一方面,智能计算框架代码迭代迅速、测试困难;另一方面,与传统软件不同,智能计算框架涉及大量张量计算,其代码规范缺乏软件工程理论指导.为了解决这一问题,现有的工作主要使用模糊测试手段实现缺陷定位,然而,这类方法只能实现特定类型缺陷的精准定位,却难以即时地在开发过程中引导开发者关注软件质量.因此,将国内外常见的智能计算框架(TensorFlow,百度飞桨等)作为研究对象,选取多种变更特征构建数据集,在代码提交级别对智能计算框架进行即时缺陷预测.另外,在此基础上使用LDA主题建模技术挖掘代码和代码提交信息作为新的特征,并使用随机森林进行预测.结果发现AUC-ROC平均值为0.77,且语义信息可以略微提升预测性能.最后,使用可解释机器学习方法 SHAP分析各特征属性对模型预测输出的影响,发现:(1)基本特征对于模型的影响符合传统软件开发规律;(2)代码和提交信息中的语义特征对模型的预测结果有重要影响;(3)不同系统中的不同特征对模型预测输出的贡献度排...     

基于语义特征抽取的文本聚类研究

基于向量空间模型(VSM)的文本聚类会出现向量维度过高以及缺乏语义信息的问题,导致聚类效果出现偏差。为解决以上问题,引入《知网》作为语义词典,并改进词语相似度算法的不足。利用改进的词语语义相似度算法对文本特征进行语义压缩,使所有特征词都是主题相关的,利用调整后的TF-IDF算法对特征项进行加权,完成文本特征抽取,降低文本表示模型的维度。在聚类中,将同一类的文本划分为同一个簇,利用簇中所有文本的特征词完成簇的语义特征抽取,簇的表示模型和文本的表示模型有着相同的形式。通过计算簇之间的语义相似度,将相似度大于阈值的簇合并,更新簇的特征,直到算法结束。通过实验验证,与基于K-Means和VSM的聚类算法相比,文中算法大幅降低了向量维度,聚类效果也有明显提升。     

语义特征造型中对象族模型研究

提出了一种新的对象族模型,即陈述式对象族模型。定义了陈述式对象族模型,给出了模型的几何结构和拓扑结构,通过约束详细说明了特征的语义,以约束图的形式给出了该模型中特征的表示,给出了模型的语法描述。通过实例验证了模型的有效性,克服了传统的基于历程建模的不足,提高了CAD建模设计效率,降低了设计成本。     

A novel framework for concept detection on large scale video database and feature pool

Large-scale semantic concept detection from large video database suffers from large variations among different semantic concepts as well as their corresponding effective low-level features. In this paper, we propose a novel framework to deal with this obstacle. The proposed framework consists of four major components: feature pool construction, pre-filtering, modeling, and classification. First, a large low-level feature pool is constructed, from which a specific set of features are selected for the latter steps automatically or semi-automatically. Then, to deal with the unbalance problem in training set, a pre-filtering classifier is generated, which the aim of achieving a high recall rate and a certain precision rate nearly 50% for a certain concept. Thereafter, from the pre-filtered training samples, a SVM classifier is built based on the selected features in the feature pool. After that, the SVM classifier is applied to classification of semantic concept. This framework is flexible and extensible in terms of adding new features into the feature pool, introducing human interactions in selecting features, building models for new concepts and adopting active learning.     

Decimation of human face model for real-time animation in intelligent multimedia systems

Animating a complex human face model in real-time is not a trivial task in intelligent multimedia systems for next generation environments. This paper proposes a generation scheme of a simplified model for real-time human face animation in intelligent multimedia systems. Previous work mainly focused on the geometric features when generating a simplified human face model. Such methods may lose the critical feature points for animating human faces. The proposed method can find those important feature points and can generate the feature-preserved low-level models busing our new quadrics. The new quadrics consist of basic error metrics and feature edge quadrics. The quality of facial animation with a lower-level model is as good as that of a computationally expansive original model. In this paper, we prove that our decimated facial model is effective in facial animation using a well-known expression-retargeting technique.     

Numeric and curve parameters for freeform surface feature models

The success of feature modelling systems depends to a large extent on offering user-friendly ways for specifying and varying feature models. In this paper, several new facilities are introduced for this.Numeric and curve parameters are presented for specification and variation of freeform surface feature models. Several properties of the parameters can be defined, such as the direction of evaluation for a curve parameter. In addition to parameters present in features, other parameters can be added to a feature model, including intersection curves between features.To provide these facilities, a new three-level structure is introduced, which includes a partially evaluated model that can capture new generic model entities, including feature intersection curves. The work has been implemented in a prototype surface feature modelling system with functionality for feature class definition, feature model creation, and advanced feature model variation through the new types of parameters.     

结合MASP和语义分割的双链路行人重识别方法

行人重识别是通过不同的摄像机识别同一个人。由于人的姿势多变,背景杂乱以及拍摄角度不同等,提取强大的行人特征成为一个有挑战性的任务。为了提取良好的行人特征表示,提出了一种结合MASP与语义分割的双链路行人重识别模型。该方法对网络不同深度的特征进行采样,不同深度的特征图具有不同的表达能力,使网络可以学习到行人身上更加细粒度的特征。上层链路针对网络过深导致行人信息丢失的问题,提出了MASP模块,对浅层特征进行采样,然后与高级特征连接,对深浅层级特征交融,增加特征的多样性。下层链路基于语义分割结果,对提取的中间层行人特征映射,得出语义部位特征。在测试阶段,将全局特征与语义部位特征结合生成多层次特征,加强模型的表征能力。在Market-1501和DukeMTMC-reID两个数据集上与其他方法的对比以及消融实验表明,提出的结合MASP与语义分割的双链路重识别模型有效提升行人重识别性能。