Explore the vast range of titles available.

Occlusion is probably the biggest challenge for human pose estimation in the wild. Typical solutions often rely on intrusive sensors such as IMUs to detect occluded joints. To make the task truly unconstrained, we present AdaFuse, an adaptive multiview fusion method, which can enhance the features in occluded views by leveraging those in visible views. The core of AdaFuse is to determine the point-point correspondence between two views which we solve effectively by exploring the sparsity of the heatmap representation. We also learn an adaptive fusion weight for each camera view to reflect its feature quality in order to reduce the chance that good features are undesirably corrupted by “bad” views. The fusion model is trained end-to-end with the pose estimation network, and can be directly applied to new camera configurations without additional adaptation. We extensively evaluate the approach on three public datasets including Human3.6M, Total Capture and CMU Panoptic. It outperforms the state-of-the-arts on all of them. We also create a large scale synthetic dataset Occlusion-Person, which allows us to perform numerical evaluation on the occluded joints, as it provides occlusion labels for every joint in the images. The dataset and code are released at https://github.com/zhezh/adafuse-3d-human-pose.

Based on the failure of the initial grouting scheme in tunnel engineering, an integrated geological forecasting system incorporating an industrial endoscope was used to detect watery karst areas in this tunnel. Numerical simulations were used to analyze the change patterns of arch top settlement, arch bottom bulge, and water surge in the tunnel under different grouting thicknesses. Compared to the displacement of the support structure, curtain grouting thickness is mainly reflected in reducing the amount of water surge. When the grouting thickness exceeds 5 m, the water-stopping effect of the tunnel is almost unchanged. Finally, a grout thickness of 5 m and a grout length of 25 m were selected as the grouting range for this project. During grouting, the combination of the three grouting techniques can effectively solve the problems of high water surges and difficult hole formation. The principle of “combination of exploration and injection” was followed to obtain real-time geological information and optimize the subsequent grouting plan. After the grouting, the grouting evaluation results and the field construction conditions showed that the grouting effect was good.

With the development of infrastructure construction in mountainous areas, the number of new extra-long tunnels is increasing. However, these tunnels often face the challenge of complex and variable surrounding rock grades, resulting in a large number of overbreak and underbreak due to the untimely adjustment of smooth blasting parameters. This study focuses on the optimization of the peripheral hole charging structure and blasting parameters for extra-long hard rock tunnels, aiming to improve the effectiveness of smooth blasting technology. The results of this study demonstrate a significant improvement in the effect of smooth blasting after implementing bidirectional polymerization blasting in the tunnel. A comparison between the bidirectional shaped charge and spaced decoupled charge blasting reveals that the former yields better results. To obtain accurate data on the tunnel section profile during excavation, a laser cross-section meter is used for measurement. Furthermore, this study quantitatively compares the optimization effect of smooth blasting parameters. The multifractal characteristics of the tunnel profile overbreak point sequences are analyzed under different smooth blasting schemes using the multifractal detrended fluctuation analysis (MF-DFA) method. It is found that both the spaced decoupled charge and the bidirectional shaped charge blasting exhibit multifractal features in the overbreak measurement point sequences. The calculation results of the multifractal features of the tunnel profile under different smooth blasting plans are in line with the actual situation.

To better analyze the fluctuation characteristics and development law of tunnel deformation data, multifractal theory is applied to tunnel deformation analysis. That is, the multifractal detrended fluctuation analysis (MF-DFA) model is first utilized to carry out the multifractal characterization of tunnel deformation data. Further, Mann–Kendall (M–K) analysis is utilized to construct the dual criterion (∆α indicator criterion and ∆f(α) indicator criterion) for the tunnel deformation early warning study. In addition, the particle swarm optimization long-short-term memory (PSO-LSTM) prediction model is used for predicting tunnel settlement. The results show that, in reference to the tunnel warning level criteria and based on the Z-value results of the indicator criterion, the warning level of all four sections is class II. At the same time, through the analysis of tunnel settlement predictions, the PSO-LSTM model has a better prediction effect and stability for tunnel settlement. The predicted results show a slow increase in tunnel settlement over the next 5 days. Finally, the tunnel warning level and the predicted results of tunnel settlement are analyzed in a comprehensive manner. The deformation will increase slowly in the future. Therefore, monitoring and measurement should be strengthened, and disaster preparedness plans should be prepared.

The demand for tunnel construction continues to grow by leaps and bounds. Therefore, tunnel mechanization construction is receiving more and more attention for improving excavation ergonomics. To enhance the scientific and comprehensive evaluation results of tunnel drilling and blasting method excavation ergonomics, a set of evaluation methods of tunnel drilling and blasting method excavation ergonomics based on the game theory G2-EW-TOPSIS model is proposed. From the three dimensions of drilling efficiency, construction process duration, and synergistic influence factors, a tunnel drilling and blasting construction ergonomics evaluation index system consisting of 11 indicators such as perimeter hole drilling efficiency, drilling duration, construction quality, and comprehensive cost is constructed. The subjective and objective weights of evaluation indicators are calculated by using the improved sequential relationship analysis method (G2 method) and entropy weight method, respectively, and the combination weights are carried out by using game theory method (GTM) with the Nash equilibrium as the goal. The indices are classified into five grades: excellent (I), good (II), average (III), rather poor (IV), and poor (V), according to the daily tunnel construction. The excavation ergonomics index to be evaluated is calculated using the combined weights, and the comprehensive evaluation index of excavation ergonomics to be evaluated is calculated using the technique for order preference by similarity to an ideal solution (TOPSIS). The proposed rating model was used to analyze the excavation ergonomics of the Shangtianling Tunnel in the Chizhou–Huangshan High-Speed Railway using jumbo drills (JD) and drilling machines (DM) in large- and small-mileage construction, respectively, and to obtain the excavation ergonomics rating and comprehensive evaluation rating of each evaluation object. The research results show that the established excavation ergonomics evaluation model can effectively identify the main factors affecting the excavation ergonomics of the drill and blast method, and has a certain reference value.

Machine-made sand is gradually replacing natural sand to achieve sustainable development. Experimental studies and gray-correlation analysis were used to study the properties of tunnel slag machine-made mortar and concrete. The properties of machine-made mortar with different stone powder content were analyzed through experiments. By analyzing the performance of machine-made sand concrete with equal amounts of cement replaced by stone powder, the optimum replacement ratio is obtained. Gray-correlation analysis was used to compare the degree of influence of fineness modulus and stone powder content on the performance of concrete. Scanning electron microscopy (SEM) and X-ray diffractometry (XRD) were used to analyze the microstructure of tunnel slag sand concrete. The test results showed that the flexural and compressive strengths of the machine-made sand concrete were greater than the standard sand with the same stone powder content. The 28-day flexural and compressive strengths had a maximum difference of more than 30%. The best stone powder content of the machine-made mortar is in the range of 5% to 8%. When the replacement cement content of stone powder is about 6%, the mechanical and working properties of machine-made sand concrete achieve the optimal state. The lower the stone powder content, the closer the mechanical and working properties of machine-made sand concrete and river sand concrete. The correlation between the performance of machine-made sand concrete and fineness modulus is the largest. When the stone powder content is low, it has almost no effect on the compressive strength of concrete. The results point out the direction for the quality control of tunnel slag machine-made sand concrete.

Computer vision researchers spent a lot of time creating large datasets, yet there is still much information that is difficult to label. Detailed annotations like part segmentation and dense keypoint are expensive to annotate. 3D information requires extra hardware to capture. Besides the labeling cost, an image dataset also lacks the ability to allow an intelligent agent to interact with the world. As a human, we learn through interaction, rather than per-pixel labeled images. To fill in the gap of existing datasets, we propose to build virtual worlds using computer graphics and use generated synthetic data to solve these challenges. In this dissertation, I demonstrate cases where computer vision challenges can be solved with synthetic data. The first part describes our engineering effort about building a simulation pipeline. The second and third part describes using synthetic data to train better models and diagnose trained models. The major challenge for using synthetic data is the domain gap between real and synthetic. In the model training part, I present two cases, which have different characteristics in terms of domain gap. Two domain adaptation methods are proposed, respectively. Synthetic data saves enormous labeling effort by providing detailed ground truth. In the model diagnosis part, I present how to control nuisance factors to analyze model robustness. Finally, I summarize future research directions that can benefit from synthetic data.

How to provide high quality data for computer vision is a challenge. Researchers spent a lot of effort creating image datasets with more images and more detailed annotation. Computer graphics (CG) is a way of creating synthetic images, during the image synthesis many types of information of the CG scene can be exported as ground truth annotation. In this paper, we develop a pipeline to synthesize realistic human images and automatically generate detailed annotation at the same time. We use 2D annotation to control the pose of the CG human model, which enables our images to contain more poses than motion capture based method. The synthetic images are used to train and evaluate human pose estimation algorithm to show its usefulness.

Occlusion is probably the biggest challenge for human pose estimation in the wild. Typical solutions often rely on intrusive sensors such as IMUs to detect occluded joints. To make the task truly unconstrained, we present AdaFuse, an adaptive multiview fusion method, which can enhance the features in occluded views by leveraging those in visible views. The core of AdaFuse is to determine the point-point correspondence between two views which we solve effectively by exploring the sparsity of the heatmap representation. We also learn an adaptive fusion weight for each camera view to reflect its feature quality in order to reduce the chance that good features are undesirably corrupted by ``bad'' views. The fusion model is trained end-to-end with the pose estimation network, and can be directly applied to new camera configurations without additional adaptation. We extensively evaluate the approach on three public datasets including Human3.6M, Total Capture and CMU Panoptic. It outperforms the state-of-the-arts on all of them. We also create a large scale synthetic dataset Occlusion-Person, which allows us to perform numerical evaluation on the occluded joints, as it provides occlusion labels for every joint in the images. The dataset and code are released at https://github.com/zhezh/adafuse-3d-human-pose.

In the realm of image generation, creating customized images from visual prompt with additional textual instruction emerges as a promising endeavor. However, existing methods, both tuning-based and tuning-free, struggle with interpreting the subject-essential attributes from the visual prompt. This leads to subject-irrelevant attributes infiltrating the generation process, ultimately compromising the personalization quality in both editability and ID preservation. In this paper, we present DisEnvisioner, a novel approach for effectively extracting and enriching the subject-essential features while filtering out -irrelevant information, enabling exceptional customization performance, in a tuning-free manner and using only a single image. Specifically, the feature of the subject and other irrelevant components are effectively separated into distinctive visual tokens, enabling a much more accurate customization. Aiming to further improving the ID consistency, we enrich the disentangled features, sculpting them into more granular representations. Experiments demonstrate the superiority of our approach over existing methods in instruction response (editability), ID consistency, inference speed, and the overall image quality, highlighting the effectiveness and efficiency of DisEnvisioner. Project page: https://disenvisioner.github.io/.