Search

The implementation of convolutional neural networks (CNNs) and Transformers has significantly accelerated the booming advances of human pose estimation. However, challenges persist in accurately estimating target details. Pose estimation faces inherent difficulties when applied in complex environments marked by intricate conditions, including factors like motion blur and chaotic scenes. In this paper, we revisit the design of CNNs and Transformers, delving deeper into their internal structures. We sequentially utilize CNNs and Transformers, leveraging the proficiency of CNNs in extracting low-level features and the capability of Transformers in establishing long-range dependencies. Building upon this framework, we introduce modifications to both CNNs and Transformer-related structures, enhancing the overall expressive capacity of the model. The modification is made to the original CNNs section: we alter BasicBlock to AtBlock to maintain high-resolution information exchange to further excavate details of objects. The two following modifications are applied to the subsequent Transformer section: (1) we replace the self-attention layer in each encoder block with the local enhancement self-attention module to capture local information. (2) We propose a local perception feed-forward network to substitute for the feed-forward network layer in each encoder block, which employs the depth-wise convolution to enhance the correlation of neighbour information in the spatial dimension. Our modifications contribute to analyzing poses that are arduous to estimate due to occlusion. Our method combines self-attention and depth-wise convolution, named CSDNet. The experiments on both COCO2017 and MPII datasets show improved performance over the baseline. Compared to other models achieving the similar accuracy, our model has fewer parameters and requires less computation. Additionally, in complicated environments such as poor lighting conditions, our method can more accurately estimate fuzzy keypoints. [ABSTRACT FROM AUTHOR]

Punching shear failure of flat slabs is a critical safety issue since it may occur without any prior signs of damage. Welded shear reinforcement fabricated by longitudinal chords and diagonal webs is proposed as a novel type of shear reinforcement to significantly improve punching shear strength and ductility of flat slabs. However, there are few investigations on its force mechanism at present. To compensate for the objective flaws in the experiment, this research uses ABAQUS finite element software to distinguish the influence of welded shear reinforcements on punching shear behavior of flat slabs. A nonlinear finite element model was developed and validated with experimental results. The failure mode, concrete stress, flexural reinforcement strain, and welded shear reinforcement strain are all investigated in failure mechanism of flat slabs. The webs of weld shear reinforcement can limit the development of cracks and strengthen the aggregate interlocking at the cracks. The parametric analysis of the welded shear reinforcement under different geometrical parameters proves that punching shear effect of diagonal webs of welded shear reinforcement is much larger than that of chords. The definition of the web shear reinforcement ratio provides a theoretical basis for the design of such punching shear keys. [ABSTRACT FROM AUTHOR]

In this paper, we are interested in the human pose estimation problem with a focus on leveraging discriminative pose features. Recent pose estimation works concentrate on extracting high-level features but ignore the low-level details, thus reducing the prediction accuracy. To mitigate the above issues, we propose an end-to-end method called multi-scale representation transformer network (MSRT). Our network consists of two key components: feature aggregation module (FAM) and transformers. The FAM splits and stacks feature maps of different scales, then fuses them to achieve multi-scale representation learning. This module makes up for the lack of detailed information in the high-level features. Furthermore, we utilize Transformers to identify long-range interactions among feature maps, and capture implicit body structure information, which allows the proposed network to refine the locations of terminal and occluded joints. Compared with existing regression-based methods, MSRT achieves superior results on the COCO2017 and MPII datasets. [ABSTRACT FROM AUTHOR]

Structures usually exhibit hysteretic and deteriorative behaviors under strong loading, which is generally modeled as a multi-degree-of-freedom nonlinear and hysteretic system under non-stationary random excitations. The resulting governing Fokker–Planck–Kolmogorov (FPK) equation with high dimensions and hysteretic nonlinearity is difficult to solve, especially when the time evolution is involved. It has been rarely analyzed. In this paper, the state-space-split exponential–polynomial closure method, previously proposed for the stationary solution of the high-dimensional FPK equation, is further generalized to consider the non-stationary effect and hysteretic nonlinearity. With the proposed solution procedure, a ten-degree-of-freedom Bouc–Wen hysteretic system under non-stationary excitation is investigated. Different shapes of hysteretic loops with softening and hardening hysteretic behaviors are considered. Accompanied system deterioration is expressed with different levels. Non-stationary responses are analyzed in terms of the expected hysteretic energy and probability density function. Pertinent Monte Carlo simulation method is performed to verify the reliability of approximate solutions. Moreover, the influence of different degradation levels on the response is discussed. [ABSTRACT FROM AUTHOR]

Accurate crowd counting is still challenging due to the variations of crowd heads. Most of crowd counting methods adopt multi-branch networks to extract multi-scale information. However, these networks are too complex to be optimized. To solve these problems, we propose an efficient scale-aware crowd counting network named SC2Net, which adopts the encoder-decoder framework. The encoder uses the first ten layers of VGG16 to extract the primary feature information. The decoder is mainly consisted of our proposed residual pyramid dilated convolution (ResPyDConv) modules to regress predicted density maps. Specifically, the ResPyDConv module is composed of pyramid dilated convolution (PyDConv). Each PyDConv adopts dilated convolutions with different dilated rates. PyDConv divides feature maps into different groups and extracts multi-scale feature information. Extensive experiments are conducted on ShanghaiTech, UCF_CC_50, UCF_QNRF, and NWPU_Crowd datasets. Qualitative and quantitive results show the superiority of our proposed network to the other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

In recent years, the researches on concrete-filled steel tubular (CFST) intersecting nodes mainly focused on the plane intersecting nodes, while that on space intersecting nodes with off-plane angles was relatively less. In this study, the nonlinear analyses of CFST space intersecting node models established in ABAQUS were performed under the monotonic axial compression and reciprocal axial tension-compression loads, respectively. The stress distributions of steel tubes and core concrete were analyzed, and the hysteresis characteristics and stiffness degradation of the node were studied. The in-plane angle, off-plane angle, out-of-plane constraint effect, steel tube diameter-to-thickness ratio, and core concrete strength were used as control parameters to analyze their effects on the mechanical behavior of the nodes. The results show that the bearing capacity of the nodes meets the design requirements, the hysteresis loop is full, the initial stiffness is large and the stiffness degrades slowly. The off-plane angle, out-of-plane constraint effect, steel tube diameter-thickness ratio and core concrete strength have significant effects on the bearing capacity and lateral stiffness of the nodes. In order to reduce the adverse effect of out-of-plane displacement, the space intersecting nodes in practical engineering can use reinforced floor beams or prestressd cables to impose out-of-plane constraint, and the recommended value of out-of-plane constraint effect is 10–30 kN/mm. [ABSTRACT FROM AUTHOR]