Group activity recognition aims to identify a consistent group from different actions performed by respective individuals. Most existing methods focus on learning the interaction between each two individuals ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e</i> ., second-order interaction). In this work, we argue that interactive relation is insufficient address task. We propose xmlns:xlink="http://www.w3.org/1999/xlink">third-order...

Weakly supervised temporal action localization (TAL) aims to localize the instances in untrimmed videos using only video-level labels. Without snippet-level labels, this task should be hard distinguish all snippets with accurate action/background categories. The main difficulties are large variations brought by unconstraint background and multiple subactions snippets. existing prototype model focuses on describing covering them clusters (defined as prototypes). In work, we argue that...

Action anticipation aims to infer the action in unobserved segment (future segment) with observed (past segment). Existing methods focus on learning key past semantics predict future, but they do not model temporal continuity between and future. However, actions are always highly uncertain anticipating The absence of smoothing video's past-and-future segments may result an inconsistent future action. In this work, we aim smooth global changes segments. We propose a Consistency-guided...

Controlling the dispersion of nanoparticles in polymer matrices is desired for nearly all applications ranging from consumer electronics to automotive tires. In nanocomposites, commonly accepted picture that individual are separated each other matrix, but this well-dispersed morphology only realized a small subset model systems. Such systems often rely on hydrophobically modified silica particles available commercial suppliers. work, we investigate how surface chemistry hydrophilic colloidal...

The causality relation modeling remains a challenging task for group activity recognition. relations describe the influence on centric actor (effect actor) from its correlative actors (cause actors). Most existing graph models focus learning with synchronous temporal features, which is insufficient to deal asynchronous features. In this paper, we propose an Actor-Centric Causality Graph Model, learns three modules, i.e., detection module, feature fusion and inference module. First, given...

Social relation, as the basic relation in our daily life, is vital for social action analysis. However, how to learn feature between people still not tackled. In this work, we propose a gaze-aware graph convolutional network (GA-GCN) recognition, which targets discovering context-aware inference with attention. To predict gaze direction, apply trained direction loss. Then, build module, two-stream both attention and distance-aware The can pick up relevant context objects representation. We...

Temporal modeling still remains as a challenge for action recognition. Most existing temporal models focus on learning local variation between neighbor frames. There exists obvious deviations and global variations, such subtle notable motion variations. In this paper, we propose difference module recognition, which consists of two sub-modules, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.</i> , aggregation module. These sub-modules...

Recently, phase-based motion estimation method is able to extract the full-field vibration of large structure from video, which has attracted widely attention. However, it suffers diverse disturbances in realistic measurement, such as periodic texture pattern surface and shake caused by unstable tripod. To address this issue, a spatiotemporal disturbance-adaptive morphological component analysis (DAMCA) proposed paper. This focuses on separating each video frame, global signal extracted...

Being lack of theoretical support from biological cues in computer vision, current computational and learning approaches object categorization mostly aim at better performances neglecting analysis on framework human brain for visual information processing materially which cause little-marginal improvement more complexity. Focusing the uncertainty color mechanism cortex motivating issues shape information, we present model incorporating invariant descriptors plausible feature biologically to...

Text segmentation is a fundamental step in natural language processing (NLP) and information retrieval (IR) tasks. Most existing approaches do not explicitly take into account the facet of documents for segmentation. annotation are often addressed as separate problems, but they operate common input space. This article proposes FTS, which novel model faceted text via multitask learning (MTL). FTS models an MTL problem with annotation. employs bidirectional long short-term memory (Bi-LSTM)...

Abstract Anomaly event detection is vital in surveillance video analysis. However, how to learn the discriminative motion crowd scene still not tackled. Here, a deep social force network by exploiting both extracting and coding proposed. Given grid of particles with velocity provided optical flow, interaction investigated module embedded network. A convolution was further designed 3D (DMC‐3D) module. The DMC‐3D only eliminates noise spatial encoder–decoder but also learns feature...

Data hiding in a cover image can be used to assist secure message communication on the Internet. In this paper, we proposed hybrid data scheme that is combination of least significant bit substitution (LSB), exploiting modification direction (EMD) and prediction errors (MPE). The aim maintain balance between embedding quality capacity where high payload motive. first embedded by EMD, or LSB followed EMD only if peak signal-to-noise ratio (PSNR) greater equal T1 dB (45 dB). remainder will...

Semantic issues are highly concerned with high-level interpretation in image understanding, which include text-image gap and its own affinity. Concentrating on text-formatting entities images, three sophisticated methodologies roundly reviewed as generative, discriminative descriptive grammar the basis of contextual features. The following objective benchmark for visual words is also directly presented semantic coherency. Finally, summarized directions semantics understanding discussed...

Monocular depth estimation is an ill-posed problem because infinite 3D scenes can be projected to the same 2D scenes. Most recent methods focus on image-level information from deep convolutional neural networks, while training them may suffer slow convergence and accuracy degeneration, especially for deeper network more feature channels. Based encoder-decoder framework, we propose a novel Residual DenseASPP Network. In our network, define features as low/mid/high vision use two-kinds of skip...

Due to the importance of feature extraction and scene representation in classification tasks, this paper presents an approach for unsupervised learning using Independent Subspace Analysis. The optimization process bases is incorporated into framework incremental cope with difficulty large or dynamic samples. proposed method could automatically learn image features accomplish Spatial Pyramid Matching model. Also, influence related parameters discussed. Experiment shows constructs efficient...

针对SAR海冰图像受相干斑噪声影响严重,提出采用相干斑抑制区域生长模型的区域MRF（SRRG-MRF）分割算法。SRRGåŒºåŸŸæ¨¡åž‹åŒ æ‹¬æž„å»ºå›¾åƒçš„ç›¸å¹²æ–‘æŠ‘åˆ¶åŒºåŸŸåŒ–è¡¨è¾¾å’ŒåŸºäºŽåŒºåŸŸçš„ç°åº¦ç›¸ä¼¼æ€§è¿›è¡ŒåŒºåŸŸç”Ÿé•¿ä¸¤ä¸ªéƒ¨åˆ†,å ¶ä¸­ç›¸å¹²æ–‘æŠ‘åˆ¶çš„åŒºåŸŸåŒ–è¡¨è¾¾ç”±ç›¸å¹²æ–‘æŠ‘åˆ¶çš„åŒè¾¹æ»¤æ³¢ï¼ˆSRBF）算法和分水岭变换构成,è¯¥æ¨¡åž‹åœ¨ç›¸å¹²æ–‘å™ªå£°ä¸¥é‡çš„æƒ å†µä¸‹,能够有效抑制过分割和对目æ...

The algorithm of causal anomaly detection in industrial control physics is proposed to determine the normal cloud line system so as accurately detect anomaly. In this paper, modeling combining Maximum Information Coefficient and Transfer Entropy was used construct network among nodes system. Then, abnormal propagation path are deduced from structural changes before after attack. Finally, an based on hybrid differential cumulative identify specific data node. stability causality mining...

This paper mainly focuses on the issues about generic multi-scale object perception for detection or recognition. A novel computational model in visually-feature space is presented scene & representation to purse underlying textural manifold statistically nonparametric manner. The associative method approximately makes perceptual hierarchy human-vision biologically coherency specific quad-tree-pyramid structure, and appropriate scale-value of different objects can automatically be selected...