Wenyuan Tao

A kind of cultural heritage, a Mani heap is piled with many carving stones and used as a religious altar for prayer in daily life in Tibet. It has distinctive characteristics and high research value, providing extensive content and abundant information. This paper focuses on an overview of three digitization projects for Mani heaps. Based on the collected data of five field surveys, the three projects leverage digitization methods to introduce, analyze, and represent Mani heaps to support scholarly analysis and casual appreciation. The authors explore these projects to study, represent, and conserve Mani heaps, which are often ignored by researchers.

Practicing gymnastics activities at home with online resources has become an increasingly popular choice due to its convenience and accessibility. However, without face-to-face guidance by a trainer, a major challenge is how to assess the quality of performed gymnastics activities, effectively and fairly. Existing intrusive assessing approaches usually require live cameras or wearable sensors, which usually generate privacy and feasibility concerns. There is a lacking of accurate approaches to assess the quality of the activities. To address these challenges, a gymnastics activity assessment approach is proposed in this article, and Wi-Gym, an effective first-of-its-kind gymnastics activity assessment system is developed utilizing commodity Wi-Fi. Wi-Gym is designed to compare the activity-induced channel state information (CSI) dynamics by an exerciser and that of a trainer utilizing dynamic time warping (DTW). The comparison results are provided by a fuzzy inference system (FIS). To make Wi-Gym robust to the changes in the environment, domain adaptation is leveraged to mitigate the data distribution imbalance caused by the environment changes. Extensive experimental studies have been conducted using Wi-Gym, acoustic, and video-based sensing systems. The experimental results validate the effectiveness and robustness of the proposed approach.

Change detection and coregistration problems, essential tasks in remote sensing image processing, can identify the differences in images of the same scene (e.g., buildings or forests) captured at two distinct time points. These images may be detected by the same or different sensors. Recently, new methods have been developed for such detection tasks, including fully convolutional networks and attention mechanisms, with favorable results. However, in fully convolutional networks, some useless information may be extracted from the multidimensional information of an image. Real-time processing is difficult because of location discrepancies, the large scale of images, and the fusion cropping problem; thus, change detection may be time-consuming and require a substantial amount of hardware resources. In our study, a novel network model was designed to overcome these problems by using high- and low-frequency image information for feature extraction. By using similarity and homography measurements, our proposed system solved the change detection and coregistration problems efficiently. A batch change detection dataset and an established discrepancy dataset were used for evaluation. The results revealed that the proposed method achieved high accuracy in detecting changes in real-time.

Multi‐dimensional data exploration is a classic research topic in visualization. Most existing approaches are designed for identifying record patterns in dimensional space or subspace. In this paper, we propose a visual analytics approach to exploring subset patterns. The core of the approach is a subset embedding network (SEN) that represents a group of subsets as uniformly‐formatted embeddings. We implement the SEN as multiple subnets with separate loss functions. The design enables to handle arbitrary subsets and capture the similarity of subsets on single features, thus achieving accurate pattern exploration, which in most cases is searching for subsets having similar values on few features. Moreover, each subnet is a fully‐connected neural network with one hidden layer. The simple structure brings high training efficiency. We integrate the SEN into a visualization system that achieves a 3‐step workflow. Specifically, analysts (1) partition the given dataset into subsets, (2) select portions in a projected latent space created using the SEN, and (3) determine the existence of patterns within selected subsets. Generally, the system combines visualizations, interactions, automatic methods, and quantitative measures to balance the exploration flexibility and operation efficiency, and improve the interpretability and faithfulness of the identified patterns. Case studies and quantitative experiments on multiple open datasets demonstrate the general applicability and effectiveness of our approach.

Analyzing the geometric and semantic properties of 3D point cloud data via the deep learning networks is still challenging due to the irregularity and sparsity of samplings of their geometric structures. In our study, the authors combine the advantage of voxels and point clouds by presenting a new data form of voxel models, called Layer‐Ring data. This data type can retain the fine description of the 3D data, and keep the high efficiency of feature extraction. After that, based on the Layer‐Ring data, a modern network architecture, called VoxPoint Annular Network (VAN), works on the Layer‐Ring data for the feature extraction and object category prediction. The design idea is based on the edge‐extraction and the coordinate representation for each voxel on the separated layer With the flexible design, our proposed VAN can adapt to the layer's geometric variability and scalability Finally, the extensive experiments and comparisons demonstrate that our approach obtained the notable results with the state‐of‐the‐art methods on a variety of standard benchmark datasets (e.g., ModelNet10, ModelNet40). Moreover, the tests also proved that 3D shape features could learn efficiently and robustly. All relevant codes will be available at https://github.com/helloFionaQ/Vox‐PointNet.

Traffic flow prediction is essential in Intelligent Transportation System (ITS). Due to the complexity of the traffic environment, predicting future traffic speed is difficult. A lot of methods have been proposed to forecast traffic flow, such as support vector regression (SVR), recurrent neural networks (RNN), and so on. However, there are few ensemble models methods for predicting traffic flow, especially for predicting multi-step traffic speed. Meanwhile, these ensemble methods often use many base models for regression. Thus, the performances of these existing methods are worth exploring for multistep traffic forecasts with few base models. Besides, this paper proposes a model based on ensemble models, called "Triplet Decoders Neural Network" (TDNN), to compare with these methods. By analyzing the characteristic of traffic speed data, this paper also proposes a data preprocessing model called "TConversion" to help RNN capturing long-term dependencies. Our experiments are implemented based on real traffic speed data from Baidu in Beijing, China because the prediction for complex traffic is more valuable. With these new novel ideas, our experiments show the superior performances of TDNN, and the significant effect of T-Conversion for predicting traffic speed sequences.

We present a novel loss function, namely, GO loss, for classification. Most of the existing methods, such as center loss and contrastive loss, dynamically determine the convergence direction of the sample features during the training process. By contrast, GO loss decomposes the convergence direction into two mutually orthogonal components, namely, tangential and radial directions, and conducts optimization on them separately. The two components theoretically affect the interclass separation and the intraclass compactness of the distribution of the sample features, respectively. Thus, separately minimizing losses on them can avoid the effects of their optimization. Accordingly, a stable convergence center can be obtained for each of them. Moreover, we assume that the two components follow Gaussian distribution, which is proved as an effective way to accurately model training features for improving the classification effects. Experiments on multiple classification benchmarks, such as MNIST, CIFAR, and ImageNet, demonstrate the effectiveness of GO loss.