Recognizing roads and intersections in aerial images is a challenging problem computer vision with many real world applications, such as localization navigation for unmanned vehicles (UAVs). The currently gaining momentum still far from being solved. While recent approaches have greatly improved due to the advances deep learning, they provide only pixel-level semantic segmentations. In this paper, we argue that should be recognized at higher level of road graphs - edges connect nodes....

Graphene and its functionalised derivatives are transforming the development of biosensors that capable detecting nucleic acid hybridization. Using a Molecular Dynamics (MD) approach, we explored single-stranded or double-stranded deoxyribose (ssDNA dsDNA) adsorption on two graphenic species: graphene oxide (GO) reduced functionalized with aminated polyethylene glycol (rGO-PEG-NH2). Innovatively, included chloride (Cl−) magnesium (Mg2+) ions influenced both ssDNA dsDNA GO rGO-PEG-NH2...

It is a truth universally acknowledged that "a picture worth thousand words". The emerge of digital media has taken this saying to complete new level. By using steganography, one can hide not only 1000, but thousands words even in an average sized image. This article presents various types techniques used by modern as well the implementation least significant bit (LSB) method. main objective develop application uses LSB insertion order encode data into cover Both serial and parallel version...

Derivatives are a crucial ingredient to broad variety of computational techniques in science and engineering. While numerical approaches for evaluating derivatives suffer from truncation error, automatic differentiation is accurate up machine precision. The term comprises set mechanically transforming given computer program another one capable derivatives. A common misconception about that this technique only works on local pieces fairly simple code. Here, it shown not applicable small...

We present a method for learning multiple scene representations given small labeled set, by exploiting the relationships between such in form of multi-task hypergraph. also show how we can use hypergraph to improve powerful pretrained VisTrans-former model without any additional data. In our hypergraph, each node is an interpretation layer (e.g., depth or segmentation) scene. Within hyperedge, one several input nodes predict at output node. Thus, could be some hyperedges and others. this...

This study investigates whether analyzing the code comments available in source can effectively reveal functional similarities within software. The authors explore how both machine-readable (such as linter instructions) and human-readable (in natural language) contribute towards measuring similarity. For former, work is relying on computing cosine similarity over one-hot encoded representation of comments, while for latter, focus detecting English using threshold-based computations against...

Semantic segmentation and vision-based geolocalization in aerial images are challenging tasks computer vision. Due to the advent of deep convolutional nets availability relatively low cost UAVs, they currently generating a growing attention field. We propose novel multi-task multi-stage neural network that is able handle two problems at same time, single forward pass. The first stage our predicts pixelwise class labels, while second provides precise location using branches. One branch uses...

Road detection from aerial images is a challenging task for humans and machines alike. Occlusion, the lack of visual cues slim class borders other road-like structures (such as pathways or private alleys) make problem inherently ambiguous, requiring logic that goes beyond input image. We propose three-stage method road segmentation - first, an ensemble multiple U-Net like CNNs generate binary masks. Second, another CNN learns to refine roads segmentations based on fusion maps first stage....

Summary Analyzing data type produced, stored, and aggregated in Big Data environments is a challenge understanding quality represents crucial support for decisionmaking. application modeling requires meta‐data modeling, interaction execution modeling. Entropy, relative entropy, mutual information play important roles theory. Our purpose within this paper to present new upper bound the classical Shannon's entropy. The derived from refinement of recent result literature, inequality S. Dragomir...

This work presents the current design and implementation of ADiJaC, an automatic differentiation tool for Java classfiles. ADiJaC uses source transformation to generate derivative codes in both forward reverse modes differentiation. We describe overall architecture present various details examples each two emphasize enhancements that have been made over previous versions illustrate their influence on generality performance generated codes. The has used derivatives a variety problems,...

Estimating precise metric depth is an essential task for UAV navigation, which very difficult to learn unsupervised without access odometry. At the same time, recovery from kinematics and optical flow mathematically precise, but less numerically stable robust, especially in focus of expansion areas. We propose a model that combines analytical, vision-with-odometry approach, with deep learning, into single formulation estimation, both fast accurate. The two pathways – analytical data-driven...

In this paper we discuss the potential of integrated GPU to accelerate sorting by performing a partial sort prior comparison based CPU sort. We experiment along with several algorithms and outline performance gain for random input data set. then analyze different ×86 SoC architectures, show that chunks stored inside onchip memory, can almost eliminate impact memory hierarchy has on performance. Finally, how our approach is from previous designs, being specifically tailored an GPU, able...

We address the challenging problem of semi-supervised learning in context multiple visual interpretations world by finding consensus a graph neural networks. Each node is scene interpretation layer, while each edge deep net that transforms one layer at into another from different node. During supervised phase networks are trained independently. next unsupervised stage nets on pseudo-ground truth provided among paths reach nets' start and end nodes. These act as ensemble teachers for any...

Energy efficiency in the buildings sector is one of main research areas on which European Union has focused its efforts. Reducing energy consumption become a priority, as well estimation and prediction. There are number different models used to achieve latter, from white-box (fully informed) black-box (all necessary information inferred data), hybrid approaches. However, these existing gray-box rely mathematical can easily be over-parameterized. We have implemented model propose an improved...

In this paper we discuss cosmological N-Body/SPH simulations on parallel computing systems with distributed memory using GADGET-2. GADGET-2 (GAlaxies Dark matter and Gas intEracT) is a novel simulation code, written in C++ publicly available, developed by Volker Springel at the Max-Planck-Institute for Astrophysics Munchen, Germany as an improved version of GADGET. It massively code that uses explicit communication model implemented standardized MPI interface. Our contribution consists...

This paper presents Reward Based Routing Protocol (RBRP), a novel routing protocol for wireless mesh networks, that aims at reducing and fairly distributing the Electromagnetic Field (EMF) Exposure caused by transmissions. The basis of this is reward-based scheme in which intermediate nodes forward packets receive reward proportional to degree exposure they generate. By means simulation campaign over NS-3 framework, we assess RBRP clearly outperforms legacy approach terms exposure, while it...