Multiview subspace clustering aims to discover the inherent structure of data by fusing multiple views complementary information. Most existing methods first extract types handcrafted features and then learn a joint affinity matrix for clustering. The disadvantage this approach lies in two aspects: 1) multiview relations are not embedded into feature learning 2) end-to-end manner deep is suitable Even when have been extracted, it nontrivial problem choose proper backbone on different...

Recently, 3D point cloud is becoming popular due to its capability represent the real world for advanced content modality in modern communication systems. In view of wide applications, especially immersive towards human perception, quality metrics clouds are essential. Existing evaluations rely on a full or certain portion original cloud, which severely limits their applications. To overcome this problem, we propose novel deep learning-based no reference assessment method, namely PQA-Net....

In modern industry, large-scale fault diagnosis of complex systems is emerging and becoming increasingly important. Most deep learning-based methods perform well on small number diagnosis, but cannot converge to satisfactory results when handling because the huge types will lead problems intra/inter-class distance unbalance poor local minima in neural networks. To address above problems, a progressive knowledge transfer-based multitask convolutional network (PKT-MCNN) proposed. First,...

Open set recognition enables deep neural networks (DNNs) to identify samples of unknown classes, while maintaining high classification accuracy on known classes. Existing methods based auto-encoder (AE) and prototype learning show great potential in handling this challenging task. In study, we propose a novel method, called Class-Specific Semantic Reconstruction (CSSR), that integrates the power AE learning. Specifically, CSSR replaces points with manifolds represented by class-specific AEs....

Multi-drone multi-target tracking aims at collabo- ratively detecting and targets across multiple drones associating the identities of objects from different drones, which can overcome shortcomings single-drone object tracking. To address critical challenges identity association target occlusion in multi-drone tasks, we collect an occlusion-aware dataset named MDMT. It contains 88 video sequences with 39,678 frames, including 11,454 IDs persons, bicycles, cars. The MDMT comprises 2,204,620...

The sizes of datasets in terms the number samples, features, and classes have dramatically increased recent years. In particular, there usually exists a hierarchical structure among class labels as hundreds exist classification task. We call these tasks classification, structures are helpful for dividing very large task into collection relatively small subtasks. Various algorithms been developed to select informative features flat classification. However, ignore semantic hyponymy directory...

Multi-view learning has progressed rapidly in recent years. Although many previous studies assume that each instance appears all views, it is common real-world applications for instances to be missing from some resulting incomplete multi-view data. To tackle this problem, we propose a novel Latent Heterogeneous Graph Network (LHGN) learning, which aims use multiple views as fully possible flexible manner. By unified latent representation, trade-off between consistency and complementarity...

Tree ensembles such as Random Forests have achieved impressive empirical success across a wide variety of applications. To understand how these models make predictions, people routinely turn to feature importance measures calculated from tree ensembles. It has long been known that Mean Decrease Impurity (MDI), one the most widely used importance, incorrectly assigns high noisy features, leading systematic bias in selection. In this paper, we address selection MDI both theoretical and...

LiDARs are widely used in autonomous robots due to their ability provide accurate environment structural information. However, the large size of point clouds poses challenges terms data storage and transmission. In this paper, we propose a novel cloud compression transmission framework for resource-constrained robotic applications, called RCPCC. We iteratively fit surface with similar range value eliminate redundancy through spatial relationships. Then, use Shape-adaptive DCT (SA-DCT)...

Open set recognition (OSR) requires models to classify known samples while detecting unknown for real-world applications. Existing studies show impressive progress using from auxiliary datasets regularize OSR models, but they have proved be sensitive selecting such outliers. In this paper, we discuss the aforementioned problem a new perspective: Can without elaborately outliers? We first empirically and theoretically explore role of foregrounds backgrounds in open disclose that: 1) that...

Attributed graph clustering aims to partition nodes of a structure into different groups. Recent works usually use variational autoencoder (VGAE) make the node representations obey specific distribution. Although they have shown promising results, how introduce supervised information guide representation learning and improve performance is still an open problem. In this article, we propose Collaborative Decision-Reinforced Self-Supervision (CDRS) method solve problem, in which pseudo...

Open set recognition requires models to recognize samples of known classes learned in the training while reject unknowns not learned. Compared with structural risk minimization theory for closed-set problems, open tasks remains rarely explored. In this paper, we point out that balancing between and space is crucial recognition, re-formalize it as risk. This brings a new view towards general relationship closed against common intuition, which argues good classifier always benefits...

Few-shot learning (FSL) is a challenging task in classifying new classes from few labelled examples. Many existing models embed class structural knowledge as prior to enhance FSL against data scarcity. However, they fall short of connecting the with limited visual information which plays decisive role model performance. In this paper, we propose unified framework multi-granularity fusion and decision-making (MGKFD) overcome limitation. We aim simultaneously explore knowledge, working mutual...

Coleoptera, including the family Nitidulidae, are valuable for estimating long-term postmortem intervals in late stage of body decomposition. This study showed that, under seven constant temperatures 16, 19, 22, 25, 28, 31, and 34 °C, developmental durations Nitidula rufipes (Linnaeus, 1767) from oviposition to eclosion were 71.0 ± 4.4, 52.9 4.1, 40.1 3.4, 30.1 2.1, 24.2 2.0, 21.0 ±2.3, 20.8 2.4 days, respectively. The morphological indexes length, widths head capsules, distance between...

Deep learning models suffer from catastrophic forgetting when new tasks incrementally. Incremental has been proposed to retain the knowledge of old classes while identify classes. A typical approach is use a few exemplars avoid knowledge. In such scenario, data imbalance between and key issue that leads performance degradation model. Several strategies have designed rectify bias towards due imbalance. However, they heavily rely on assumptions relation Therefore, are not suitable for complex...

Collaborative applications of physical systems and algorithms bring the rapid development cyber (CPS). Establishing CPS with image classification systems, however, is difficult, because both categories algorithms, deep learning methods traditional feature extraction methods, are independent individual currently. Therefore, in this paper, we propose a fast fusion algorithm to satisfy requirement area from comprehensive perspective. First, fuse shallow-layer network feature, large pre-trained...

Hierarchical structures of labels usually exist in large-scale classification tasks, where can be organized into a tree-shaped structure. The nodes near the root stand for coarser labels, while close to leaves mean finer labels. We label unseen samples from node leaf node, and obtain multigranularity predictions hierarchical classification. Sometimes, we cannot decision due uncertainty or incomplete information. In this case, should stop at an internal rather than going ahead rashly....

In large-scale data classification tasks, it is becoming more and challenging in finding a true class from huge amount of candidate categories. Fortunately, hierarchical structure usually exists these massive The task utilizing this for effective called classification. It follows top-down fashion which predicts sample the root node with coarse-grained category to leaf fine-grained category. However, misclassification inevitable if information insufficient or large uncertainty prediction...