Search

Abstract Few-shot Semantic Segmentation (FSS) was proposed to segment unseen classes in a query image, referring to only a few annotated examples named support images. One of the characteristics of FSS is spatial inconsistency between query and support targets, e.g., texture or appearance. This greatly challenges the generalization ability of methods for FSS, which requires to effectively exploit the dependency of the query image and the support examples. Most existing methods abstracted support features into prototype vectors and implemented the interaction with query features using cosine similarity or feature concatenation. However, this simple interaction may not capture spatial details in query features. To address this limitation, some methods utilized pixel-level support information by computing pixel-level correlations between paired query and support features implemented with the attention mechanism of Transformer. Nevertheless, these approaches suffer from heavy computation due to dot-product attention between all pixels of support and query features. In this paper, we propose a novel framework, termed ProtoFormer, built upon the Transformer architecture, to fully capture spatial details in query features. ProtoFormer treats the abstracted prototype of the target class in support features as the Query and the query features as Key and Value embeddings, which are input to the Transformer decoder. This approach enables better capture of spatial details and focuses on the semantic features of the target class in the query image. The output of the Transformer-based module can be interpreted as semantic-aware dynamic kernels that filter the segmentation mask from the enriched query features. Extensive experiments conducted on PASCAL- $$5^{i}$$ 5 i and COCO- $$20^{i}$$ 20 i datasets demonstrate that ProtoFormer significantly outperforms the state-of-the-art methods in FSS.

Abstract Inpainting high-resolution images with large holes challenges existing deep learning-based image inpainting methods. We present a novel framework—PyramidFill for high-resolution image inpainting, which explicitly disentangles the task into two sub-tasks: content completion and texture synthesis. PyramidFill attempts to complete the content of unknown regions in a lower-resolution image, and synthesize the textures of unknown regions in a higher-resolution image, progressively. Thus, our model consists of a pyramid of fully convolutional GANs, wherein the content GAN is responsible for completing contents in the lowest-resolution masked image, and each texture GAN is responsible for synthesizing textures in a higher-resolution image. Since completing contents and synthesizing textures demand different abilities from generators, we customize different architectures for the content GAN and texture GAN. Experiments on multiple datasets including CelebA-HQ, Places2 and a new natural scenery dataset (NSHQ) with different resolutions demonstrate that PyramidFill generates higher-quality inpainting results than the state-of-the-art methods.

This study investigates the relationship between eye-tracking metrics and emotional experiences in the context of cultural landscapes and tourism-related visual stimuli. Fifty-three participants were involved in two experiments: forty-three in the data collection phase and ten in the model validation phase. Eye movements were recorded and the data were analyzed to identify correlations between four eye-tracking metrics—average number of saccades (ANS), total dwell fixation (TDF), fixation count (FC), and average pupil dilation (APD)—and 19 distinct emotional experiences, which were subsequently grouped into three categories: positive, neutral, and negative. The study examined the variations in eye-tracking metrics across architectural, historic, economic, and life landscapes, as well as the three primary phases of a tour: entry, core, and departure. Findings revealed that architectural and historic landscapes demanded higher levels of visual and cognitive engagement, especially during the core phase. Stepwise regression analysis identified four key eye-tracking predictors for emotional experiences, enabling the development of a prediction model. This research underscores the effectiveness of eye-tracking technology in capturing and predicting emotional responses to different landscape types, offering valuable insights for optimizing rural tourism environments and enhancing visitors’ emotional experiences.

In this paper, the wave characteristics of true surface waves (TSWs) and pseudo surface waves (PSWs) in two-dimensional (2D) seismic metamaterials are explored. The 2D seismic metamaterial is formed by placing periodic concrete pillars on the surface of the soil substrate. The complex dispersion curves of the seismic metamaterial are numerically calculated by the ω–k approach. The Kelvin–Voigt model is adopted to describe the material viscoelasticity. The results show that not only the TSW modes but also the PSW modes should be taken into account to evaluate the mitigation capacity of the seismic metamaterials. Due to the introduction of material viscoelasticity, the TSWs and PSWs convert from the propagative mode to the evanescent mode. When the PSW modes propagate along the substrate surface, their energy gradually leaks into the bulk of the substrate, leading to their transmission coefficient smaller than that of the TSW modes. Compared to the concrete pillars on the substrate surface, the viscoelasticity of the soil substrate plays a more important role in surface wave reduction. Moreover, the wave propagating length in the seismic metamaterials can be predicted by using complex dispersion relations.

BackgroundSchool bullying may cause sleep disorders in early adolescents. Here, we determined the relationship between school bullying (considering all the features of bullying involvement) and sleep disorders, which are the common problems in Chinese early adolescents.Materials and methodsWe conducted a questionnaire survey among 5,724 middle school students from Xuancheng, Hefei, and Huaibei cities in Anhui province, China. The self-report questionnaires included the Olweus Bully/Victim Questionnaire and Pittsburgh Sleep Quality Index. We used latent class analysis to identify the potential subgroups of bullying behavior. Logistic regression analysis was used to investigate the association between school bullying and sleep disorders.ResultsActive participants in bullying interactions, including the bullies and victims, reported higher levels of sleep disorders compared with the non-active participants [Bully: physical (aOR = 2.62), verbal (aOR = 1.73), relational (aOR = 1.80), and cyber (aOR = 2.08); Victim: physical (aOR = 2.42), verbal (aOR = 2.59), relational (aOR = 2.61), and cyber (aOR = 2.81)]. A dose–response relationship was observed between the number of school bullying types and sleep disorders. In the context of bullying roles, bully-victims had the highest risk of reporting sleep disorders (aOR = 3.07, 95% CI: 2.55–3.69). We identified four potential categories of school bullying behaviors: low involvement in bullying, verbal and relational victims, medium bully-victims, and high bully-victims, and the highest frequency of sleep disorders was observed in the high bully-victims group (aOR = 4.12, 95% CI: 2.94–5.76).ConclusionOur findings indicate a positive correlation between bullying roles and sleep disorders in early adolescents. Therefore, targeted intervention for sleep disorders should include an evaluation of bullying experiences.

Abstract Aims Previous cross-sectional and case–control studies have proposed that decreased vitamin D levels are positively correlated with the risk of suicidality in adults. However, limited studies have examined the association between vitamin D and suicidality in adolescents. This study aimed to investigate the relationship between serum vitamin D and suicidality risk among early adolescents. Methods Data were obtained from a Chinese early adolescent cohort. In this cohort, seventh-grade students from a middle school in Anhui Province were invited to voluntarily participate in the baseline assessments and provide peripheral blood samples (in September 2019). The participants were followed up annually (in September 2020 and September 2021). Serum 25-hydroxyvitamin D [25(OH)D] and vitamin D–related single-nucleotide polymorphisms at baseline were measured in November 2021. Traditional observational and Mendelian randomization (MR) analyses were performed to examine the relationship between serum 25(OH)D at baseline and the risk of baseline and incident suicidality (i.e., suicidal ideation [SI], plans and attempts). Results Traditional observational analysis did not reveal a significant linear or non-linear association of serum 25(OH)D concentration with the risks of baseline and 2-year incident suicidality in the total sample (P > .05 for all). Sex-stratified analysis revealed a non-linear association between the 25(OH)D concentration and the risk of baseline SI in women (Poverall = .002; Pnon-linear = .001). Moreover, the risk of baseline SI in the 25(OH) insufficiency group was lower than that in the 25(OH) deficiency group in the total sample (odds ratio [OR] = 0.69, 95% confidence interval [CI] = 0.51–0.92, P = .012). This difference remained significant in women (OR = 0.59, 95% CI = 0.40–0.87, P = .008) but not in men (OR = 0.78, 95% CI = 0.53–1.15, P = .205). Additionally, both linear and non-linear MR analyses did not support the causal effect of serum 25(OH)D concentration on the risk of baseline, 1-year and 2-year incident suicidality (P > .05 for all). Conclusions This study could not confirm the causal effect of vitamin D on suicidality risk among Chinese early adolescents. Future studies must confirm these findings with a large sample size.

Cervical cancer (CESC) is the fourth most common and death-causing gynecological cancer, mostly induced by infection of human papillomavirus (HPV). Multiple components of the tumor microenvironment (TME), such as tumor infiltrating immune cells, could be targets of immunotherapy for HPV-related CESC. However, little is known about the TME of CESC until now. Here, we aimed to uncover the pathogenesis as well as to identify novel biomarkers to predict prognosis and immunotherapy efficacy for CESC. Combining the transcriptomic data and clinical characteristics, we identified differentially expressed genes in CESC samples from TCGA database by comparing the two groups with different ImmuneScore and StromalScore. Next, we detected ten key genes based on the PPI network and survival analyses with the univariate Cox regression model. Thereafter, we focused on CD3G, the only gene exhibiting increased RNA and protein expression in tumors by multiple analyses. Higher CD3G expression was associated with better survival; and it was also significantly associated with immune-related pathways through GSEA analysis. Furthermore, we found that CD3G expression was correlated with 16 types of TICs. Single cell RNA-sequencing data of CD3G in lymphocytes subgroup indicated its possible role in HPV defense. Hence, CD3G might be a novel biomarker in prognosis and immunotherapy for CESC patients.

Face Parsing aims to partition the face into different semantic parts, which can be applied into many downstream tasks, e.g., face mask up, face swapping, and face animation. With the popularity of cameras, it is easier to acquire facial images. However, pixel-wise manually labeling is time-consuming and labor-intensive, which motivates us to explore the unlabeled data. In this paper, we present a self-supervised learning method attempting to make full use of the unlabeled facial images for face parsing. In particular, we randomly mask some patches in the central area of facial images, and the model is required to reconstruct the masked patches. This self-supervised pretraining is capable of making the model capture facial feature representations through these unlabeled data. After self-supervised pretraining, the model is fine-tuned on a few labeled data for the face parsing task. Experimental results show that the model achieves better performance for face parsing assisted by the self-supervised pretraining, which greatly decreases the labeling cost. Our approach achieves 74.41 mIoU on the LaPa test set fine-tuned on only 0.2% of the labeled data of the whole training data, surpassing the model that is directly trained by a large margin of +5.02 mIoU. In addition, our approach achieves a new state-of-the-art on the LaPa and CelebAMask-HQ test set.

It is difficult to have enough samples to implement the full-scale life test on the loader drive axle due to high cost. But the extreme small sample size can hardly meet the statistical requirements of the traditional reliability analysis methods. In this work, the method of combining virtual sample expanding with Bootstrap is proposed to evaluate the fatigue reliability of the loader drive axle with extreme small sample. First, the sample size is expanded by virtual augmentation method to meet the requirement of Bootstrap method. Then, a modified Bootstrap method is used to evaluate the fatigue reliability of the expanded sample. Finally, the feasibility and reliability of the method are verified by comparing the results with the semi-empirical estimation method. Moreover, from the practical perspective, the promising result from this study indicates that the proposed method is more efficient than the semi-empirical method. The proposed method provides a new way for the reliability evaluation of costly and complex structures.

This article focuses on the shift strategy of hydro-mechanical infinitely variable transmission. A computer simulation is carried out to investigate the shift strategy, and the simulation results are verified by the test data. There are four typical working conditions in the continuous shift process, among which the shift process from F 2 (N) to F 3 (P) plays a decisive role in shift quality. The shift strategy based on the physical parameters from F 2 (N) to F 3 (P) is lower engine rotating speed, smaller load torque, lower main circuit pressure, and larger speed valve flow rate. The shift strategy based on the shift time from F 2 (N) to F 3 (P) is to switch the third group of clutches first, then disengage the first group of clutches, and engage the second group of clutches at last. The results show that the shift quality of hydro-mechanical infinitely variable transmission can be improved greatly by the optimization of physical parameters and shift time.