scholarly journals WS-AM: Weakly Supervised Attention Map for Scene Recognition

Electronics ◽  
2019 ◽  
Vol 8 (10) ◽  
pp. 1072 ◽  
Author(s):  
Shifeng Xia ◽  
Jiexian Zeng ◽  
Lu Leng ◽  
Xiang Fu
Keyword(s):  
Large Scale ◽  
Multiple Scales ◽  
Scene Recognition ◽  
Small Scale ◽  
Great Success ◽  
Global Feature ◽  
Benchmark Datasets ◽  
Local Mean ◽  
Weakly Supervised ◽  
Fully Connected

Recently, convolutional neural networks (CNNs) have achieved great success in scene recognition. Compared with traditional hand-crafted features, CNN can be used to extract more robust and generalized features for scene recognition. However, the existing scene recognition methods based on CNN do not sufficiently take into account the relationship between image regions and categories when choosing local regions, which results in many redundant local regions and degrades recognition accuracy. In this paper, we propose an effective method for exploring discriminative regions of the scene image. Our method utilizes the gradient-weighted class activation mapping (Grad-CAM) technique and weakly supervised information to generate the attention map (AM) of scene images, dubbed WS-AM—weakly supervised attention map. The regions, where the local mean and the local center value are both large in the AM, correspond to the discriminative regions helpful for scene recognition. We sampled discriminative regions on multiple scales and extracted the features of large-scale and small-scale regions with two different pre-trained CNNs, respectively. The features from two different scales were aggregated by the improved vector of locally aggregated descriptor (VLAD) coding and max pooling, respectively. Finally, the pre-trained CNN was used to extract the global feature of the image in the fully- connected (fc) layer, and the local features were combined with the global feature to obtain the image representation. We validated the effectiveness of our method on three benchmark datasets: MIT Indoor 67, Scene 15, and UIUC Sports, and obtained 85.67%, 94.80%, and 95.12% accuracy, respectively. Compared with some state-of-the-art methods, the WS-AM method requires fewer local regions, so it has a better real-time performance.

Geomechanical and Fracture Network Interpretation of a Devonian Outcrop

2021 ◽  
pp. 1-16
Author(s):  
Scott McKean ◽  
Simon Poirier ◽  
Henry Galvis-Portilla ◽  
Marco Venieri ◽  
Jeffrey A. Priest ◽  
...  
Keyword(s):  
Large Scale ◽  
Multiple Scales ◽  
Fracture Network ◽  
Natural Fracture ◽  
Small Scale ◽  
Nugget Effect ◽  
Schmidt Hammer ◽  
Unconventional Reservoir ◽  
Fracture Anisotropy ◽  
Fracture Intensity

Summary The Duvernay Formation is an unconventional reservoir characterized by induced seismicity and fluid migration, with natural fractures likely contributing to both cases. An alpine outcrop of the Perdrix and Flume formations, correlative with the subsurface Duvernay and Waterways formations, was investigated to characterize natural fracture networks. A semiautomated image-segmentation and fracture analysis was applied to orthomosaics generated from a photogrammetric survey to assess small- and large-scale fracture intensity and rock mass heterogeneity. The study also included manual scanlines, fracture windows, and Schmidt hammer measurements. The Perdrix section transitions from brittle fractures to en echelon fractures and shear-damage zones. Multiple scales of fractures were observed, including unconfined, bedbound fractures, and fold-relatedbed-parallel partings (BPPs). Variograms indicate a significant nugget effect along with fracture anisotropy. Schmidt hammer results lack correlation with fracture intensity. The Flume pavements exhibit a regionally extensive perpendicular joint set, tectonically driven fracturing, and multiple fault-damage zones with subvertical fractures dominating. Similar to the Perdrix, variograms show a significant nugget effect, highlighting fracture anisotropy. The results from this study suggest that small-scale fractures are inherently stochastic and that fractures observed at core scale should not be extrapolated to represent large-scale fracture systems; instead, the effects of small-scale fractures are best represented using an effective continuum approach. In contrast, large-scale fractures are more predictable according to structural setting and should be characterized robustly using geological principles. This study is especially applicable for operators and regulators in the Duvernay and similar formations where unconventional reservoir units abut carbonate formations.

Three-dimensional conditional structure of a high-Reynolds-number turbulent boundary layer

2011 ◽  
Vol 673 ◽  
pp. 255-285 ◽  
Author(s):  
N. HUTCHINS ◽  
J. P. MONTY ◽  
B. GANAPATHISUBRAMANI ◽  
H. C. H. NG ◽  
I. MARUSIC
Keyword(s):  
Skin Friction ◽  
High Reynolds Number ◽  
Large Scale ◽  
Three Dimensional ◽  
Streamwise Velocity ◽  
Small Scale ◽  
Large Scale Structures ◽  
Local Mean ◽  
Scale Structures ◽  
High Reynolds

An array of surface hot-film shear-stress sensors together with a traversing hot-wire probe is used to identify the conditional structure associated with a large-scale skin-friction event in a high-Reynolds-number turbulent boundary layer. It is found that the large-scale skin-friction events convect at a velocity that is much faster than the local mean in the near-wall region (the convection velocity for large-scale skin-friction fluctuations is found to be close to the local mean at the midpoint of the logarithmic region). Instantaneous shear-stress data indicate the presence of large-scale structures at the wall that are comparable in scale and arrangement to the superstructure events that have been previously observed to populate the logarithmic regions of turbulent boundary layers. Conditional averages of streamwise velocity computed based on a low skin-friction footprint at the wall offer a wider three-dimensional view of the average superstructure event. These events consist of highly elongated forward-leaning low-speed structures, flanked on either side by high-speed events of similar general form. An analysis of small-scale energy associated with these large-scale events reveals that the small-scale velocity fluctuations are attenuated near the wall and upstream of a low skin-friction event, while downstream and above the low skin-friction event, the fluctuations are significantly amplified. In general, it is observed that the attenuation and amplification of the small-scale energy seems to approximately align with large-scale regions of streamwise acceleration and deceleration, respectively. Further conditional averaging based on streamwise skin-friction gradients confirms this observation. A conditioning scheme to detect the presence of meandering large-scale structures is also proposed. The large-scale meandering events are shown to be a possible source of the strong streamwise velocity gradients, and as such play a significant role in modulating the small-scale motions.

scholarly journals Assemblages and Scale in Archaeology

2017 ◽  
Vol 27 (1) ◽  
pp. 127-139 ◽  
Author(s):  
Oliver J.T. Harris
Keyword(s):  
Large Scale ◽  
Multiple Scales ◽  
Daily Life ◽  
Critical Issue ◽  
Small Scale ◽  
Processes Of Change ◽  
Origins Of Agriculture ◽  
The Real ◽  
Historical Processes

The growing interest in assemblages has already opened up a number of important lines of enquiry in archaeology, from the morphogenetic capacities of matter through to a rethinking of the concept of community. In this paper I want to explore how assemblages allow us to reconceptualize the critical issue of scale. Archaeologists have vacillated between expending energy on the ‘great processes’ of change like the evolution of humanity, the colonization of the globe or the origins of agriculture, and focusing on the momentary, fleeting nature of a small-scale ethnographic present. Where archaeologists have attempted to integrate different scales the result has usually been to turn to Annales-influenced or time perspectivism-driven approaches and their fixed, linear and ontologically incompatible layers of history. In contrast, I will use assemblages to examine how we can rethink both the emergence of multiple scales and their role in history, without reducing the differences of the small-scale to an epiphenomenal outcome of larger events, or treating large-scale historical processes as mere reifications of the ‘real’ on-the-ground stuff of daily life. As we will see, this approach also has consequences for the particular kind of reality we accord to large-scale archaeological categories.

scholarly journals The Aesthetics of Making

2021 ◽  
Author(s):  
◽  
Matt Buttimore
Keyword(s):  
Architectural Design ◽  
Large Scale ◽  
Multiple Scales ◽  
Small Scale ◽  
Urban Context ◽  
Production Scale ◽  
Community Culture ◽  
Detail Design ◽  
The People ◽  
Cultural Implications

<p><b>As the architectural design process evolves and embraces new techniques and technologies and mass production is more readily available, the relationship between designer and craftsman has become more distant. As we look to produce more and more architecture every year on a larger production scale, the craft and detail of the architecture begin to fall at the wayside. As we lose this relationship, the culture and identity of a place are also lost as these technologies are not responding to specific site and cultural implications.</b></p> <p>One such site where this is applicable is the small coastal town of Onemana in the Coromandel, a town of slightly more than 300 homes constructed as a single development in the 1980s. The rush to produce more homes and on a larger scale has meant the town’s architecture does not reflect the community culture or coastal identity of the place or the people who live there. </p> <p>This thesis argues that there is an existing relationship between craftsperson and designer and explores how this relationship and detail design can generate and inform architectural design. Understanding this relationship will generate detail design that has a more powerful outcome on the spatial qualities of the architecture and generates my own detail design language. It also argues that there exists a relationship between detail design and the urban environment, which is not fully utilised in the industry.</p> <p>The thesis proposes that this can be achieved by testing and evaluating this hypothesis across three scales and three types of urban context. The three test sites identified are a small scale private dwelling, a mid-scale cultural installation and a large scale town centre. Using the process of beginning with detail design, architectural installations will be implemented and evaluated before moving to the following location. As result the method will be proven to work across multiple scales and reflect a variety of cultural inputs.</p>

scholarly journals Blocking-based Neighbor Sampling for Large-scale Graph Neural Networks

2021 ◽  
Author(s):  
Kai-Lang Yao ◽  
Wu-Jun Li
Keyword(s):  
Neural Networks ◽  
Large Scale ◽  
Sampling Strategy ◽  
Unbiased Estimation ◽  
Small Scale ◽  
Original Graph ◽  
Exponential Increase ◽  
Benchmark Datasets ◽  
Graph Neural Networks ◽  
Memory Complexity

The exponential increase in computation and memory complexity with the depth of network has become the main impediment to the successful application of graph neural networks (GNNs) on large-scale graphs like graphs with hundreds of millions of nodes. In this paper, we propose a novel neighbor sampling strategy, dubbed blocking-based neighbor sampling (BNS), for efficient training of GNNs on large-scale graphs. Specifically, BNS adopts a policy to stochastically block the ongoing expansion of neighboring nodes, which can reduce the rate of the exponential increase in computation and memory complexity of GNNs. Furthermore, a reweighted policy is applied to graph convolution, to adjust the contribution of blocked and non-blocked neighbors to central nodes. We theoretically prove that BNS provides an unbiased estimation for the original graph convolution operation. Extensive experiments on three benchmark datasets show that, on large-scale graphs, BNS is 2X~5X faster than state-of-the-art methods when achieving the same accuracy. Moreover, even on the small-scale graphs, BNS also demonstrates the advantage of low time cost.

scholarly journals Polygon-Net: A General Framework for Jointly Boosting Multiple Unsupervised Neural Machine Translation Models

2019 ◽  
Author(s):  
Chang Xu ◽  
Tao Qin ◽  
Gang Wang ◽  
Tie-Yan Liu
Keyword(s):  
Machine Translation ◽  
Loss Function ◽  
General Framework ◽  
Large Scale ◽  
Great Success ◽  
Neural Machine Translation ◽  
Low Resource ◽  
Parallel Data ◽  
Benchmark Datasets ◽  
First Time

Neural machine translation (NMT) has achieved great success. However, collecting large-scale parallel data for training is costly and laborious.  Recently, unsupervised neural machine translation has attracted more and more attention, due to its demand for monolingual corpus only, which is common and easy to obtain, and its great potentials for the low-resource or even zero-resource machine translation. In this work, we propose a general framework called Polygon-Net, which leverages multi auxiliary languages for jointly boosting unsupervised neural machine translation models. Specifically, we design a novel loss function for multi-language unsupervised neural machine translation. In addition, different from the literature that just updating one or two models individually, Polygon-Net enables multiple unsupervised models in the framework to update in turn and enhance each other for the first time. In this way, multiple unsupervised translation models are associated with each other for training to achieve better performance. Experiments on the benchmark datasets including UN Corpus and WMT show that our approach significantly improves over the two-language based methods, and achieves better performance with more languages introduced to the framework. 

Some comparisons between heterogeneous and homogeneous layers for nonlinear SH waves in terms of heterogeneous and nonlinear effects

2020 ◽  
pp. 108128652094635 ◽  
Author(s):  
Dilek Demirkuş
Keyword(s):  
Solitary Wave ◽  
Large Scale ◽  
Multiple Scales ◽  
Nonlinear Effects ◽  
Small Scale ◽  
Solitary Wave Solutions ◽  
Nls Equation ◽  
Sh Waves ◽  
Wave Solutions ◽  
The One

This paper aims to make some comparative studies between heterogeneous and homogeneous layers for nonlinear shear horizontal (SH) waves in terms of the heterogeneous and nonlinear effects. Therefore, with this aim, two layers are defined as follows: on the one hand, one layer consists of hyperelastic, isotropic, heterogeneous, and generalized neo-Hookean materials; on the other hand, another layer is made up of hyperelastic, isotropic, homogeneous, and generalized neo-Hookean materials. Moreover, it is assumed that upper boundaries are stress-free and lower boundaries are rigidly fixed. The method of multiple scales is used in both analyses, in addition to using the known solutions of the nonlinear Schrödinger (NLS) equation, called bright and dark solitary wave solutions; these comparisons are made, numerically, and then all results are given for the lowest branch of both dispersion relations, graphically. Moreover, these comparisons are observed both on a large scale and on a small scale, not only in terms of the bright and dark solitary wave solutions but also in terms of the heterogeneous and nonlinear effects.

Use of High-Resolution DIDSON Sonar to Quantify Attributes of Predation at Ecologically Relevant Space and Time Scales

2013 ◽  
Vol 47 (1) ◽  
pp. 33-46 ◽  
Author(s):  
Victoria E. Price ◽  
Peter J. Auster ◽  
Laura Kracker
Keyword(s):  
High Resolution ◽  
Time Scales ◽  
Large Scale ◽  
Multiple Scales ◽  
Predation Rate ◽  
Visual Observation ◽  
Small Scale ◽  
Dual Frequency ◽  
Predator Prey ◽  
Space And Time

AbstractPredator-prey interactions of large vagile fishes are difficult to study in the ocean due to limitations in the space and time requirements for observations. Small-scale direct underwater observations by divers (ca. <10 m radius) and large-scale hydroacoustic surveys (10 s m2 to 100 s km2) are traditional approaches for surveying fish. However, large piscivorous predators identify and attack prey at the scale of meters to tens of meters. Dual-Frequency Identification Sonar (or DIDSON) is a high-resolution acoustic camera operating in the MHz range that provides detailed continuous video-like imaging of objects up to a range of 30 m. This technology can be used to observe predator-prey interactions at ecologically relevant space and time scales often missed by traditional methods. Here we establish an approach for quantifying predation-related behaviors from DIDSON records. Metrics related to predator and prey group size, prey responses to predation, predation rate, predator strategies, and the nonrandom use of landscape features by both predator and prey are described. In addition, relationships between patterns in these attributes are tested and issues regarding sampling strategies for future studies are discussed. We suggest that approaches combining direct visual observation and acoustic sampling at multiple scales are required to quantify variation in these relationships across underwater landscapes.

scholarly journals ActivityNet-QA: A Dataset for Understanding Complex Web Videos via Question Answering

2019 ◽  
Vol 33 ◽  
pp. 9127-9134 ◽  
Author(s):  
Zhou Yu ◽  
Dejing Xu ◽  
Jun Yu ◽  
Ting Yu ◽  
Zhou Zhao ◽  
...  
Keyword(s):  
Large Scale ◽  
Question Answering ◽  
Research Direction ◽  
Small Scale ◽  
Modeling Language ◽  
Image Domain ◽  
Recent Developments ◽  
Benchmark Datasets ◽  
Language And Vision ◽  
Web Videos

Recent developments in modeling language and vision have been successfully applied to image question answering. It is both crucial and natural to extend this research direction to the video domain for video question answering (VideoQA). Compared to the image domain where large scale and fully annotated benchmark datasets exists, VideoQA datasets are limited to small scale and are automatically generated, etc. These limitations restrict their applicability in practice. Here we introduce ActivityNet-QA, a fully annotated and large scale VideoQA dataset. The dataset consists of 58,000 QA pairs on 5,800 complex web videos derived from the popular ActivityNet dataset. We present a statistical analysis of our ActivityNet-QA dataset and conduct extensive experiments on it by comparing existing VideoQA baselines. Moreover, we explore various video representation strategies to improve VideoQA performance, especially for long videos.

scholarly journals A visual study of turbulent shear flow

1973 ◽  
Vol 61 (3) ◽  
pp. 513-540 ◽  
Author(s):  
Stavros G. Nychas ◽  
Harry C. Hershey ◽  
Robert S. Brodkey
Keyword(s):  
High Speed ◽  
Large Scale ◽  
Outer Region ◽  
Solid Particles ◽  
Turbulent Shear ◽  
Small Scale ◽  
Wall Region ◽  
Wall Area ◽  
Sequence Of Events ◽  
Local Mean

The outer region of a turbulent boundary layer along a flat plate was photographed and analysed; in addition, limited observations of the wall area were also made. The technique involved suspending very small solid particles in water and photographing their motion with a high-speed camera moving with the flow.The single most important event observed in the outer region was fluid motion which in the convected view of the travelling camera appeared as a transverse vortex. This was a large-scale motion transported downstream almost parallel to the wall with an average velocity slightly smaller than the local mean. It appeared to be the result of an instability interaction between accelerated and decelerated fluid, and it is believed to be closely associated with the wall-region ejections. The transverse vortex was part of a deterministic sequence of events; although these events occurred randomly in space and time. The first of these events was a decelerated flow exhibiting velocities considerably smaller than the local mean. It was immediately followed by an accelerated flow. Both these events extended from near the wall to the far outer region. Their interaction resulted in the formation of one or more transverse vortices. While the transverse vortex was transported downstream, small-scale fluid elements, originating in the wall area of the decelerated flow, were ejected outwards (ejection event). After travelling some distance outwards the ejected elements interacted with the oncoming accelerated fluid in the wall region and were subsequently swept downstream (sweep event). The sequence of events closed with two large-scale motions.Estimated positive and negative contributions to the instantaneous Reynolds stress during the events were many times higher than the local mean values.

Sign in / Sign up

Export Citation Format

Share Document