Object Granularity Research Articles

The Establishment of a precise intelligent evaluation system for sports events: Diving

The introduction of action quality assessment technology in sports events to achieve precise intelligent evaluation can greatly enhance the objectivity and effectiveness of competition results. Taking diving as the specific application background, this study proposes a novel Multi-granularity Extraction Approach for Temporal-spatial features in judge scoring prediction (MEAT) under the conditions of action quality assessment. On the one hand, it uses dual-modal inflated 3D ConvNet to extract the temporal and spatial features of each modal diving video at the video granularity parallelly and to merge them to form a global feature. On the other hand, the human body pose is modeled, and the simulated athlete's three-dimensional splash state is taken as local characteristics at the object granularity. Finally, the global and local features are concatenated into the fully connected layer, and heuristic method inspired by competition rules using labeled distribution learning are employed to output the probability distribution of the average score of all referees. The maximum probability score is selected and multiplied by the difficulty coefficient to obtain the final diving score. Through comprehensive experiments, comparing the Spearman's rank correlation (SRC) evaluation results of existing methods on the UNIV-Dive dataset, this framework reflects the greater accuracy advantage and further lays the foundation for the actual implementation of the technology.

Efficient Parallel Functional Programming with Effects

Although functional programming languages simplify writing safe parallel programs by helping programmers to avoid data races, they have traditionally delivered poor performance. Recent work improved performance by using a hierarchical memory architecture that allows processors to allocate and reclaim memory independently without any synchronization, solving thus the key performance challenge afflicting functional programs. The approach, however, restricts mutation, or memory effects, so as to ensure "disentanglement", a low-level memory property that guarantees independence between different heaps in the hierarchy. This paper proposes techniques for supporting entanglement and for allowing functional programs to use mutation at will. Our techniques manage entanglement by distinguishing between disentangled and entangled objects and shielding disentangled objects from the cost of entanglement management. We present a semantics that formalizes entanglement as a property at the granularity of memory objects, and define several cost metrics to reason about and bound the time and space cost of entanglement. We present an implementation of the techniques by extending the MPL compiler for Parallel ML. The extended compiler supports all features of the Parallel ML language, including unrestricted effects. Our experiments using a variety of benchmarks show that MPL incurs a small time and space overhead compared to sequential runs, scales well, and is competitive with languages such as C++, Go, Java, OCaml. These results show that our techniques can marry the safety benefits of functional programming with performance.

Weakly supervised learning of multi-object 3D scene decompositions using deep shape priors

Representing scenes at the granularity of objects is a prerequisite for scene understanding and decision making. We propose PriSMONet, a novel approach based on Prior Shape knowledge for learning Multi-Object 3D scene decomposition and representations from single images. Our approach learns to decompose images of synthetic scenes with multiple objects on a planar surface into its constituent scene objects and to infer their 3D properties from a single view. A recurrent encoder regresses a latent representation of 3D shape, pose and texture of each object from an input RGB image. By differentiable rendering, we train our model to decompose scenes from RGB-D images in a self-supervised way. The 3D shapes are represented continuously in function-space as signed distance functions which we pre-train from example shapes in a supervised way. These shape priors provide weak supervision signals to better condition the challenging overall learning task. We evaluate the accuracy of our model in inferring 3D scene layout, demonstrate its generative capabilities, assess its generalization to real images, and point out benefits of the learned representation.

Using data-mining techniques to improve combinatorial optimization algorithms

In this work, we show how data-mining can be used to cluster algorithmic generated data and use that data to improve algorithms that solve combinatorial optimization problems for a real-world application—the field-programmable gate array placement problem. Our methodology is a means for other algorithm engineers to improve their own algorithms for specific real-world problems that are hard to improve. In our case, the placement algorithms are difficult to improve, and to find better heuristics we analyze the results of placement solutions to find clustered information which can then be used to improve the algorithms. Specifically, we show a technique for gathering cluster information about placement, we create a new simulated annealing algorithm and a new genetic algorithm that can deal with a mixed granularity of placement objects on a virtual field-programmable gate array, and we show that these algorithms either execute faster or improve the overall quality of solution compared to their basic algorithm without this clustering data and improved heuristics. For our improved simulated annealing placer we improve the algorithms run-time by 17% across a range of benchmarks, and our genetic algorithm improves placement metrics—-critical path by 10% and channel-width by 4%.

Animal Species Recognition System using Deep Learning

Animals watching is a common hobby but to identify their species requires the assistance of Animal books. To provide Animal watchers a handy tool to admire the beauty of Animals, we developed a deep learning platform to assist users in recognizing species of Animals endemic to using app named the Imagenet of Animals (IoA). Animal images were learned by a convolutional neural network (CNN) to localize prominent features in the images. First, we established and generated a bounded region of interest to the shapes and colors of the object granularities and subsequently balanced the distribution of Animals species. Then, a skip connection method was used to linearly combine the outputs of the previous and current layers to improve feature extraction. Finally, we applied the SoftMax function to obtain a probability distribution of Animals features. The learned parameters of Animals features were used to identify pictures uploaded by mobile users. The proposed CNN model with skip connections achieved higher accuracy of 99.00 % compared with the 93.98% from a CNN and 89.00% from the SVM for the training images. As for the test dataset, the average sensitivity, specificity, and accuracy were 93.79%, 96.11%, and 95.37%, respectively.

Support vector machines with the known feature-evolution priors

In traditional SVMs (support vector machines), each feature involved in an object is assumed to contribute equally to the classification accuracy. However, the equal importance assumption for each feature is inappropriate for classifying the objects. The existing weighting strategies usually utilize the fixed scaling factors to measure the importance of each feature and hence they are not good fits for feature evolution. In this paper, with the assumption that a Known Feature-Evolution Prior (KFEP) can be employed to capture importance distribution from the perspective of both feature granularity and object granularity, SVM-based machines called KFEP-SVMs are developed to capture the importance of features dynamically. Also, it is revealed that SVMs with the known feature-evolution priors are equivalent to SVMs on a higher feature space. Besides, we theoretically demonstrate that the proposed machines KFEP-SVMs with the known feature-evolution prior indeed have enhanced generalization ability, compared with the classic SVMs. Experimental results show the effectiveness of the proposed machines KFEP-SVMs.

An MRF-Based Multigranularity Edge-Preservation Optimization for Semantic Segmentation of Remote Sensing Images

Semantic segmentation is one of the most important tasks in the field of remote sensing image processing. Many methods have been proposed to realize it at the pixel granularity or object granularity. Specifically, the pixel-based methods usually can effectively extract the detailed information and edges, and the object-based methods can keep the internal consistency of each land cover or land use. The Markov random field (MRF) model provides a statistical way to combine the advantages of both pixel and object granularities together. However, current MRF-based methods still face a problem, that is, how to ensure that the advantages of different granularities will complement each other, not that disadvantages will affect advantages. To solve this problem, a new multigranularity edge-preservation optimization is proposed in this letter. The proposed method first represents the image with a series of granularities from the object to the pixel by downsampling. Then, the MRF model is defined on each granularity. By defining an edge set for each granularity, during the process of downsampling, the proposed method can continuously correct edges while maintaining intraclass consistency. Experiments of Gaofen-2 and SPOT5 demonstrate the effectiveness of the proposed method. Moreover, the proposed method can be also used as the postprocessing step for deep learning. The experiment of the Pavia University hyperspectral image illustrates it for an instance of DeepLab v3+.

Rethinking safe consistency in distributed object-oriented programming

Large scale distributed systems require to embrace the trade off between consistency and availability, accepting lower levels of consistency to guarantee higher availability. Existing programming languages are, however, agnostic to this compromise, resulting in consistency guarantees that are the same for the whole application and are implicitly adopted from the middleware or hardcoded in configuration files. In this paper, we propose to integrate availability in the design of an object-oriented language, allowing developers to specify different consistency and isolation constraints in the same application at the granularity of single objects. We investigate how availability levels interact with object structure and define a type system that preserves correct program behavior. Our evaluation shows that our solution performs efficiently and improves the design of distributed applications.

Multigranularity Multiclass-Layer Markov Random Field Model for Semantic Segmentation of Remote Sensing Images

Semantic segmentation is one of the most important tasks in remote sensing. However, as spatial resolution increases, distinguishing the homogeneity of each land class and the heterogeneity between different land classes are challenging. The Markov random field model (MRF) is a widely used method for semantic segmentation due to its effective spatial context description. To improve segmentation accuracy, some MRF-based methods extract more image information by constructing the probability graph with pixel or object granularity units, and some other methods interpret the image from different semantic perspectives by building multilayer semantic classes. However, these MRF-based methods fail to capture the relationship between different granularity features extracted from the image and hierarchical semantic classes that need to be interpreted. In this article, a new MRF-based method is proposed to incorporate the multigranularity information and the multilayer semantic classes together for semantic segmentation of remote sensing images. The proposed method develops a framework that builds a hybrid probability graph on both pixel and object granularities and defines a multiclass-layer label field with hierarchical semantic over the hybrid probability graph. A generative alternating granularity inference is suggested to provide the result by iteratively passing and updating information between different granularities and hierarchical semantics. The proposed method is tested on texture images, different remote sensing images obtained by the SPOT5, Gaofen-2, GeoEye, and aerial sensors, and Pavia University hyperspectral image. Experiments demonstrate that the proposed method shows a better segmentation performance than other state-of-the-art methods.

AndroidOff:Offloading android application based on cost estimation

Computation offloading is a promising way of improving the performance and reducing the battery power consumption, since it moves some time-consuming computation activities to nearby servers. Although various approaches have been proposed to support computation offloading, we argue that there is still sufficient space for improvements, since existing approaches cannot accurately estimate the execution costs. As a result, we find that their offloading plans are less optimized. To handle the problem, in this paper, given an Android application, we propose a novel approach, called AndroidOff, that supports offloading at the granularity of objects. Supporting such capability is challenging due to the two reasons: (1) through dynamic execution, it is feasible to collect the execution costs of only partial methods, and (2) it is difficult to accurately estimate the execution costs of the remaining methods. To overcome the challenges, given an Android application, AndroidOff first combines static and dynamic analysis to predict the execution costs of all its methods. After all the costs are estimated, AndroidOff synthesizes an offloading plan, in which determines the offloading details. We evaluate AndroidOff on a real-world application, with two mobile devices. Our results show that, compared with other approaches, AndroidOff saves the response time by 8%–49% and reduces the energy consumption by 12%–49% on average for computation-intensive applications.

A Hybrid Markov Random Field Model With Multi-Granularity Information for Semantic Segmentation of Remote Sensing Imagery

High-spatial-resolution (HSR) remote sensing images usually contain rich hierarchical semantic information. However, many methods fail to solve the segmentation misclassification problems for HSR images due to just considering one layer of granularity information, such as the pixel granularity layer or the object granularity layer. This paper presents a hybrid Markov random field model for the semantic segmentation of HSR images by paying closer attention to capture the multi-granularity information. In this model, a probability graph with a multilayer structure is first built to represent different granularities information. Then, a hybrid label field is developed to model the multi-granularity classes with the form of a vector, and a new joint distribution is designed to capture the isotropic spatial interactions within each layer of the hybrid label field and the anisotropic spatial interactions between different layers. A generative probabilistic inference is proposed to realize the synergy between the multi-granularity information and the hybrid label interactions by iteratively updating the likelihood function and the joint probability of the hybrid label. The final semantic segmentation result can be achieved when the probabilistic inference converges. Experimental results over different HSR remote sensing images show that the proposed method can achieve more accurate segmentation than other state-of-the-art methods.

Software-defined object storage in multi-tenant environments

Abstract Object stores are becoming pervasive due to their scalability and simplicityto manage data growth. Their rampant adoption, however, contrasts with their scant flexibility to support multi-tenancy. Very often, this results in a deficient adaptation of the system to the heterogeneous tenants’ demands and the multiple applications sharing the same object storage infrastructure. In this manuscript, we present Crystal, the first Software-Defined Storage (SDS) architecture whose core objective is to efficiently support multi-tenancy in object stores. Abide by the principles of software-defined systems, Crystal separates the storage’s control logic from the underlying storage nodes, proxies and caches, promoting the (logical) centralization of storage control and the ability to program the storage layer. For storage control, Crystal enables high-level policies at the tenant, container, and object granularities. It also translates these policies into a set of distributed controllers that can orchestrate computations at the storage layer based on real-time workload information. We demonstrate Crystal through three use cases on top of OpenStack Swift: One that proves its storage automation capabilities; one that exploits the pushdown of lambda functions on data streams to improve analytics data ingestion; and another that differentiates IO bandwidth in a multi-tenant scenario. We show that Crystal is an extensible platform to deploy new SDS services for object stores with small overhead.

Bird Image Retrieval and Recognition Using a Deep Learning Platform

Birdwatching is a common hobby but to identify their species requires the assistance of bird books. To provide birdwatchers a handy tool to admire the beauty of birds, we developed a deep learning platform to assist users in recognizing 27 species of birds endemic to Taiwan using a mobile app named the Internet of Birds (IoB). Bird images were learned by a convolutional neural network (CNN) to localize prominent features in the images. First, we established and generated a bounded region of interest to refine the shapes and colors of the object granularities and subsequently balanced the distribution of bird species. Then, a skip connection method was used to linearly combine the outputs of the previous and current layers to improve feature extraction. Finally, we applied the softmax function to obtain a probability distribution of bird features. The learned parameters of bird features were used to identify pictures uploaded by mobile users. The proposed CNN model with skip connections achieved higher accuracy of 99.00 % compared with the 93.98% from a CNN and 89.00% from the SVM for the training images. As for the test dataset, the average sensitivity, specificity, and accuracy were 93.79%, 96.11%, and 95.37%, respectively.

Conflict analysis and detection based on model checking for spatial access control policy

In this paper, we propose a Multi-granularity Spatial Access Control (MSAC) model, in which multigranularity spatial objects introduce more types of policy rule conflicts than single-granularity objects do. To analyze and detect these conflicts, we first analyze the conflict types with respect to the relationship among the policy rules, and then formalize the conflicts by template matrices. We designed a model-checking algorithm to detect potential conflicts by establishing formalized matrices of the policy set. Lastly, we conducted experiments to verify the performance of the algorithm using various spatial data sets and rule sets. The results show that the algorithm can detect all the formalized conflicts. Moreover, the algorithm’s efficiency is more influenced by the spatial object granularity than the size of the rule set.

\u201cHigh-Throughput Characterization of Region-Specific Mitochondrial Function and Morphology\u201d

The tissue-specific etiology of aging and stress has been elusive due to limitations in data processing of current techniques. Despite that many techniques are high-throughput, they usually use singular features of the data (e.g. whole fluorescence). One technology at the nexus of fluorescence-based screens is large particle flow cytometry (“biosorter”), capable of recording positional fluorescence and object granularity information from many individual live animals. Current processing of biosorter data, however, do not integrate positional information into their analysis and data visualization. Here, we present a bioanalytical platform for the quantification of positional information (“longitudinal profiling”) of C. elegans, which we posit embodies the benefits of both high-throughput screening and high-resolution microscopy. We show the use of these techniques in (1) characterizing distinct responses of a transcriptional reporter to various stresses in defined anatomical regions, (2) identifying regions of high mitochondrial membrane potential in live animals, (3) monitoring regional mitochondrial activity in aging models and during development, and (4) screening for regulators of muscle mitochondrial dynamics in a high-throughput format. This platform offers a significant improvement in the quality of high-throughput biosorter data analysis and visualization, opening new options for region-specific phenotypic screening of complex physiological phenomena and mitochondrial biology.

Object-stack: An object-oriented approach for top-k keyword querying over fuzzy XML

Keyword search is the most popular technique of searching information from XML (eXtensible markup language) document. It enables users to easily access XML data without learning the structure query language or studying the complex data schemas. Existing traditional keyword query methods are mainly based on LCA (lowest common ancestor) semantics, in which the returned results match all keywords at the granularity of elements. In many practical applications, information is often uncertain and vague. As a result, how to identify useful information from fuzzy data is becoming an important research topic. In this paper, we focus on the issue of keyword querying on fuzzy XML data at the granularity of objects. By introducing the concept of "object tree", we propose the query semantics for keyword query at object-level. We find the minimum whole matching result object trees which contain all keywords and the partial matching result object trees which contain partial keywords, and return the root nodes of these result object trees as query results. For effectively and accurately identifying the top-K answers with the highest scores, we propose a score mechanism with the consideration of tf*idf document relevance, users' preference and possibilities of results. We propose a stack-based algorithm named object-stack to obtain the top-K answers with the highest scores. Experimental results show that the object-stack algorithm outperforms the traditional XML keyword query algorithms significantly, and it can get high quality of query results with high search efficiency on the fuzzy XML document.

Analysis of Document Summarization and Word Classification in a Smart Environment

Background/Objectives: The need for digital learning is rapidly increasing with the effective granularity of learning objects. Retrieval of learning materials from smart environment helps the learner to have a personalized experience using classification algorithms. Methods/Statistical Analysis: The digital learning objects are stored in a local repository in the smart architecture. The framework involves mobile agents which play a vital role in predicting the intuitive nature of the learner which helps the learner to get the targeted content. Content retrieval is based on the document summarisation and word classification which is proposed using classification algorithms like content similarity algorithm. Effective personalization is obtained by better accuracy in terms of precision and recall. Findings: Document summarization is done using classification algorithm initially based on the type of document retrieved by the learner. The word classification based on the retrieved document is further processed with a pre processing method which is used in the domain ontology. The similarity of the topical terms retrieved helps in classifying the document and presenting the content to the learner for more personalized experience. Further these contents which are stored in a smart framework include mobile agents which help in providing synchronous communication between the users working on similar problems at the same time and gives uninterrupted content to the user based on the query even while the user is moving in different locations. The content similarity algorithm provides better accuracy of retrieved data thereby providing the exact document requested by the learner. The samples of the data classification provide the expected the result of the retrieved data which involve document summarization based on the word classification. The results have proved that the learner has a better intuitive experience with the targeted document wherein the learner gets better personalization. Applications/Improvements: More intelligent agents can be involved for learner as well as author to make the system more personalized and provide better collaboration. The classification algorithms along with clustering of documents obtained after summarization can yield better accuracy for large data sets.

Scale problems in assessment of hydrogeological parameters of groundwater flow models

Abstract An overview is presented of scale problems in groundwater flow, with emphasis on upscaling of hydraulic conductivity, being a brief summary of the conventional upscaling approach with some attention paid to recently emerged approaches. The focus is on essential aspects which may be an advantage in comparison to the occasionally extremely extensive summaries presented in the literature. In the present paper the concept of scale is introduced as an indispensable part of system analysis applied to hydrogeology. The concept is illustrated with a simple hydrogeological system for which definitions of four major ingredients of scale are presented: (i) spatial extent and geometry of hydrogeological system, (ii) spatial continuity and granularity of both natural and man-made objects within the system, (iii) duration of the system and (iv) continuity/granularity of natural and man-related variables of groundwater flow system. Scales used in hydrogeology are categorised into five classes: micro-scale – scale of pores, meso-scale – scale of laboratory sample, macro-scale – scale of typical blocks in numerical models of groundwater flow, local-scale – scale of an aquifer/aquitard and regional-scale – scale of series of aquifers and aquitards. Variables, parameters and groundwater flow equations for the three lowest scales, i.e., pore-scale, sample-scale and (numerical) block-scale, are discussed in detail, with the aim to justify physically deterministic procedures of upscaling from finer to coarser scales (stochastic issues of upscaling are not discussed here). Since the procedure of transition from sample-scale to block-scale is physically well based, it is a good candidate for upscaling block-scale models to local-scale models and likewise for upscaling local-scale models to regional-scale models. Also the latest results in downscaling from block-scale to sample scale are briefly referred to.

Multi-granularity spatial-temporal access control model for web GIS

The multi-granularity spatial-temporal-related access control (MSTAC) model was proposed to meet the spatial access control requirements for the service-oriented spatial data infrastructure (SDI). MSTAC extends the attribute constraints of role-based access control (RBAC), which includes the user's location attribute, the role's time constraint, the layer vector constraint of a map class, the scale and time constraints of a geographic layer, the topological constraints of geographic features, the semantic attribute expression constraints of geographic features, and the field constraint of feature views. Through this model, authorized users would be limited to access different granularity spatial datasets, such as the map granularity, the graphic layer granularity, the feature object granularity and the feature view granularity. Finally, the MSTAC model is achieved in a web GIS, which shows the positive and negative authorizations to different services in different data granularities and time periods.

CMaps, Learning Objects and a Blended Learning Signals & Systems Course

Learning Objects (LOs) are one of building blocks of Information and Communication Technology (ICT) supported learning. ICT support can be used in the traditional face-to-face model, in a completely online situation or in a hybrid (blended) style. 
 
 The development of LOs is time and resources consuming, for this reason, it is important that they can de shared and reused. This paper addresses the use o Concept Maps (CMaps) to model topics that make up the area known as Linear Time-Invariant Systems (LTI Systems) as taught in some engineering curricula. The CMaps are used to define the granularity of Learning Objects (LOs) in this area. 
 
 It also addresses the deployment of a Blended Learning Signals and Systems course that is mandatory in five undergaduate curricula. The CMaps are used to define â??regionsâ? of related LOs that are used in the guide (story board) of this course. The LOs are among the digital resources to be used.
 
 The results of this work allow future extensions in terms of examining the syllabi of the undergraduate courses that teach LTI Systems and of developing additional blended learning courses. Results are presented and future developments are also considered.