Temporal action localization (TAL) in untrimmed videos recently receives tremendous research enthusiasm. To our best knowledge, this is the first attempt literature to explore task under an unsupervised setting, hereafter referred as co-localization (ACL), where only total count of unique actions that appear video set known. solve ACL, we propose a two-step ``clustering + localization" iterative procedure. The clustering step provides noisy pseudo-labels for step, and temporal co-attention...

Video visual relation detection (VidVRD) aims to describe all interacting objects in a video. Different from relationships static images, videos contain an addition temporal channel. A majority of existing works divide video into short segments, predict each segment, and merge them. Such methods cannot capture relations involving long motions. Predicting the same relationship across neighboring segments is also inefficient. To address these issues, this work proposes novel sliding-window...

Temporal action localization is a recently-emerging task, aiming to localize video segments from untrimmed videos which contain specific actions. This work proposes novel integrated temporal scale aggregation network (TSA-Net). Our main insight that ensembling convolution filters with different dilation rates can effectively enlarge the receptive field low computational cost, inspires us devise multi-dilation (MDC) block. Furthermore, tackle instances durations, TSA-Net consists of multiple...

Multi-robot aggregation is an important application for emergent robotic tasks, in which multiple robots are aggregated to work collaboratively. In this context, the collision-free problem poses a significant challenge, complicated resolve, as prone collision. This study attempts use edge intelligence technology solve problem. First, objective function built multi-robot problem, considering characteristics of and constraints. Second, heuristic artificial plant community algorithm proposed...

Mobile Edge Computing (MEC) migrates the computing center to network edge provide services for Users (MUs). However, due limited capacity of MUs and insufficient types stored by servers, it brings great challenges computation offloading service caching in Internet Things (IoT) network. In this paper, we investigate joint optimization problem offloading, resource allocation replacement a collaborative MEC system minimize total cost all under latency constraints. Accordingly, formulate as...

With the widespread use of autonomous guided vehicles (AGVs), avoiding collisions has become a challenging problem. Addressing issue is not straightforward since production efficiency, collision avoidance, and energy consumption are conflicting factors. This paper proposes novel edge computing method based on vehicle intelligence to solve energy-efficient collision-free machine/AGV scheduling First, architecture was built, corresponding state transition diagrams for were developed. Second,...

Aiming at the problems of low detection accuracy and slow speed in white porcelain wine bottle flaw detection, an improved algorithm based on YOLOv4 was proposed. By adding Coordinate Attention to backbone feature extraction network, extracting ability features improved. Deformable convolution is added locate flaws more accurately, so as improve by model. Efficient Intersection over Union used replace Complete loss function model accuracy. Experimental results surface data set bottles show...

Temporal sentence grounding in videos is a crucial task vision-language learning. Its goal retrieving video segment from an untrimmed that semantically corresponds to natural language query. A usually contains multiple semantic events, which are rarely isolated. They tend be temporally ordered and correlated ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e.g.</i> , some event often the precursor of another event). To precisely localize...

Weakly-supervised temporal action localization aims to locate intervals of instances with only video-level labels for training. However, the results generated from video classification networks are often not accurate due lack boundary annotation actions. Our motivating insight is that should be stably predicted under various transforms. This inspires a self-supervised equivariant transform consistency constraint. We design set operations, including naive down-sampling learnable...

Turbo equalization algorithms have been used to effectively improve the performance in MIMO OFDM systems. To estimated accuracies of channel matrix, an iterative estimation using soft decided signals and pilots was proposed. Furthermore, a new method based on subtracting interference terms, it is maximum posterior probability (MAP) MMSE algorithm has lower complexity than traditional algorithms. Simulation results indicate that proposed methods can yield remarkable improvements after no less...

After more than 30 years of development, combinatorial testing has become an essential method in the field software testing, which always been active field. Through combined coverage, can detect faults caused by various parameters and their interactions a under test. This paper aims to review development briefly introduce its basic applications. We classifies today's applications, including those traditional industry IT industry. The research-and-application progress was investigated with...

Temporal action localization is a recently-emerging task, aiming to localize video segments from untrimmed videos that contain specific actions. Despite the remarkable recent progress, most two-stage methods still suffer imprecise temporal boundaries of proposals. This work proposes novel integrated scale aggregation network (TSA-Net). Our main insight ensembling convolution filters with different dilation rates can effectively enlarge receptive field low computational cost, which inspires...

Intensity inhomogeneity is one of the major obstacles for intensity‐based segmentation in many applications. The recently proposed kernel mapping (KM) method has exhibited excellent performance on segmenting various types noisy images while it not effective to handle intensity inhomogeneity. To overcome this drawback, study presents a localised KM (LKM) based fact that can be ignored local neighbourhood. authors’ first reconstructs formulation image neighbourhood each pixel, and then such...

Energy-efficient human–robot collaboration poses significant challenges to the sustainable operation of production systems. Therefore, our work proposes novel robotic edge intelligence address issue. First, is proposed fully utilize embedded computing capabilities robots, and state transition diagrams are developed for jobs, humans, respectively. Second, a multi-objective model designed energy-efficient scheduling problem evaluate performance energy efficiency as whole. Third, heuristic...

Fault localization is the critical but most expensive step in testing manufacturing software, effectively locating faults has become an increasingly concerned study. The existing spectrum-based fault techniques utilize spectrum information and specific prioritization algorithm to gener ate suspiciousness as well ranking of statements. However, effectiveness software would be dramatically reduced once statement involving bug assigned with same other non-faulty A multi-technique fusion...

How to use natural language generate program code directly has become the focus of researchers, and realizing machine's understanding is an important foundation solve this problem. In processing system based on deep learning, word embedding model can encode words sentences, which significantly improves capability a neural network process text data. Through analysis model, paper describes its vital role in generation, elaborates using automatically code, finally looks forward development trend.

Intensity inhomogeneity and noise are two major obstacles for segmenting medical images. The global kernel mapping based piecewise constant model (PCM) has superior performance on resisting noise, though it fails to cope with intensity inhomogeneity. In order overcome the difficulty caused by inhomogeneity, we first establish an energy in a neighborhood of pixel. Then such energies all pixels image integrated formulate local PCM. Energy minimization been implemented level set framework....

By analyzing the characteristic of fast fading channels, a Kalman equalization method directly estimating channel values in time domain is proposed based on synchronization orthogonal frequency division multiplexing (TDS-OFDM) system models. The algorithm one-dimensional and has low complexity. In addition, for tracking time-varying channel, minimum mean-squared error (MMSE) equalizer iterative decision-feedback are also proposed. For given signal-to-noise ratios (SNRs), simulation results...

Objective: to study the calibration period of main motor pulmonary function instrument sensor. Methods: A matched control group was used, one calibrated periodically and other not calibrated. The values oxygen sensor carbondioxide were compared. Results: electrochemical type most sensitive change time environment, carbon dioxide infrared more environment. Conclusion: sensors should be before each use.