Kaikki aineistot
Lisää
Abstract Face alignment is a crucial component in most face analysis systems. It focuses on identifying the location of several keypoints of the human faces in images or videos. Although several methods and models are available to developers in popular computer vision libraries, they still struggle with challenges such as insufficient illumination, extreme head poses, or occlusions, especially when they are constrained by the needs of real-time applications. Throughout this article, we propose a set of training strategies and implementations based on data augmentation, software optimization techniques that help in improving a large variety of models belonging to several real-time algorithms for face alignment. We propose an extended set of evaluation metrics that allow novel evaluations to mitigate the typical problems found in real-time tracking contexts. The experimental results show that the generated models using our proposed techniques are faster, smaller, more accurate, more robust in specific challenging conditions and smoother in tracking systems. In addition, the training strategy shows to be applicable across different types of devices and algorithms, making them versatile in both academic and industrial uses.
Abstract Face alignment is a crucial step in multiple face analysis and recognition tasks. The current state-of-the-art is comprised by very slow methods based on deep learning that require computationally heavy inference and very fast methods based on cascades of regressors that lack the ability to cope with complicated cases or extreme poses. The authors show how collecting a small subset of unlabelled domain-specific data can improve the accuracy of fast-inference models utilising data annotated by a slower one and a teacher–student architecture. In the proposed solution, they annotate a small subset of facial images belonging to two challenging domains using a slow but more accurate model, and this data is used to incrementally train a fast one. Their results show that by adding as little as a 5% of challenging data, they can reduce the error rate in a specific domain up to 30% without losing any generalisation abilities. This training scheme has applicability in numerous computer vision and engineering problems where computational power and model size are constrained by the application and platform or real-time operation is a requirement.
Abstract Depression is a mental illness that may be harmful to an individual’s health. The detection of mental health disorders in the early stages and a precise diagnosis are critical to avoid social, physiological, or psychological side effects. This work analyzes physiological signals to observe if different depressive states have a noticeable impact on the blood volume pulse (BVP) and the heart rate variability (HRV) response. Although typically, HRV features are calculated from biosignals obtained with contact-based sensors such as wearables, we propose instead a novel scheme that directly extracts them from facial videos, just based on visual information, removing the need for any contact-based device. Our solution is based on a pipeline that is able to extract complete remote photoplethysmography signals (rPPG) in a fully unsupervised manner. We use these rPPG signals to calculate over 60 statistical, geometrical, and physiological features that are further used to train several machine learning regressors to recognize different levels of depression. Experiments on two benchmark datasets indicate that this approach offers comparable results to other audiovisual modalities based on voice or facial expression, potentially complementing them. In addition, the results achieved for the proposed method show promising and solid performance that outperforms hand-engineered methods and is comparable to deep learning-based approaches.
Abstract We explore the possibility of leveraging radar-based sensing systems to analyze vital signs for classification, user identification, and regression tasks. Specifically, we extract time-domain and frequency-domain features from distance, respiration, and pulse signals obtained by filtering radio-frequency signals. Our Random Forest classification models are trained on these features to recognize scenarios in which the radar data were collected, categorize individuals into age groups, and classify human activities. For classification, we achieved up to 94.7% of accuracy when distinguishing apnea and normal breathing in the lying position. We then show the feasibility of identifying individuals in a small group using vital signs, which can support model fine-tuning with data acquired from new users. Furthermore, we used a Random Forest regression model to estimate the Body Mass Index, height, and weight of subjects. These classification, identification, and regression models benefit smart systems that can simultaneously identify users, recognize their behaviours, and extract their vital signs from radar sensors.
Abstract Meditation is a practice that aims at self-inducing a state of calmed rest. In this work, we analyze physiological signals collected with wearable sensors to observe if meditation has a noticeable effect on the human body response and if this effect is inversely related to stress and can be detected using the same biosignals and similar features and methods. Our work is based on the extraction of statistical and physiological features and extends the models found in the literature by extracting 30 additional features related to heart rate variability. The results show that using wrist wearable devices, meditation periods can be distinguished from spontaneous rest with an accuracy of up to 86% accuracy.
Abstract Human Activity Recognition (HAR) from wearable sensor data identifies movements or activities in unconstrained environments. HAR is a challenging problem as it presents great variability across subjects. Obtaining large amounts of labelled data is not straightforward, since wearable sensor signals are not easy to label upon simple human inspection. In our work, we propose the use of neural networks for the generation of realistic signals and features using human activity monocular videos. We show how these generated features and signals can be utilized, instead of their real counterparts, to train HAR models that can recognize activities using signals obtained with wearable sensors. To prove the validity of our methods, we perform experiments on an activity recognition dataset created for the improvement of industrial work safety. We show that our model is able to realistically generate virtual sensor signals and features usable to train a HAR classifier with comparable performance as the one trained using real sensor data. Our results enable the use of available, labeled video data for training HAR models to classify signals from wearable sensors.
Abstract Reproductive character displacement occurs when competition for successful breeding imposes a divergent selection on the interacting species, causing a divergence of reproductive traits. Here, we show that a disputed butterfly taxon is actually a case of male wing colour shift, apparently produced by reproductive character displacement. Using double digest restriction‐site associated DNA sequencing and mitochondrial DNA sequencing we studied four butterfly taxa of the subgenus Cupido (Lepidoptera: Lycaenidae): Cupido minimus and the taxon carswelli, both characterized by brown males and females, plus C. lorquinii and C. osiris, both with blue males and brown females. Unexpectedly, taxa carswelli and C. lorquinii were close to indistinguishable based on our genomic and mitochondrial data, despite displaying strikingly different male coloration. In addition, we report and analysed a brown male within the C. lorquinii range, which demonstrates that the brown morph occurs at very low frequency in C. lorquinii. Such evidence strongly suggests that carswelli is conspecific with C. lorquinii and represents populations with a fixed male brown colour morph. Considering that these brown populations occur in sympatry with or very close to the blue C. osiris, and that the blue C. lorquinii populations never do, we propose that the taxon carswelli could have lost the blue colour due to reproductive character displacement with C. osiris. Since male colour is important for conspecific recognition during courtship, we hypothesize that the observed colour shift may eventually trigger incipient speciation between blue and brown populations. Male colour seems to be an evolutionarily labile character in the Polyommatinae, and the mechanism described here might be at work in the wide diversification of this subfamily of butterflies.
Abstract Motivation: Assessing biodiversity status and trends in plant communities is critical for understanding, quantifying and predicting the effects of global change on ecosystems. Vegetation plots record the occurrence or abundance of all plant species co-occurring within delimited local areas. This allows species absences to be inferred, information seldom provided by existing global plant datasets. Although many vegetation plots have been recorded, most are not available to the global research community. A recent initiative, called ‘sPlot’, compiled the first global vegetation plot database, and continues to grow and curate it. The sPlot database, however, is extremely unbalanced spatially and environmentally, and is not open-access. Here, we address both these issues by (a) resampling the vegetation plots using several environmental variables as sampling strata and (b) securing permission from data holders of 105 local-to-regional datasets to openly release data. We thus present sPlotOpen, the largest open-access dataset of vegetation plots ever released. sPlotOpen can be used to explore global diversity at the plant community level, as ground truth data in remote sensing applications, or as a baseline for biodiversity monitoring. Main types of variable contained: Vegetation plots (n = 95,104) recording cover or abundance of naturally co-occurring vascular plant species within delimited areas. sPlotOpen contains three partially overlapping resampled datasets (c. 50,000 plots each), to be used as replicates in global analyses. Besides geographical location, date, plot size, biome, elevation, slope, aspect, vegetation type, naturalness, coverage of various vegetation layers, and source dataset, plot-level data also include community-weighted means and variances of 18 plant functional traits from the TRY Plant Trait Database. Spatial location and grain: Global, 0.01–40,000 m2. Time period and grain: 1888–2015, recording dates. Major taxa and level of measurement: 42,677 vascular plant taxa, plot-level records. Software format: Three main matrices (.csv), relationally linked.
Abstract Aims: Vegetation‐plot records provide information on the presence and cover or abundance of plants co‐occurring in the same community. Vegetation‐plot data are spread across research groups, environmental agencies and biodiversity research centers and, thus, are rarely accessible at continental or global scales. Here we present the sPlot database, which collates vegetation plots worldwide to allow for the exploration of global patterns in taxonomic, functional and phylogenetic diversity at the plant community level. Results: sPlot version 2.1 contains records from 1,121,244 vegetation plots, which comprise 23,586,216 records of plant species and their relative cover or abundance in plots collected worldwide between 1885 and 2015. We complemented the information for each plot by retrieving climate and soil conditions and the biogeographic context (e.g., biomes) from external sources, and by calculating community‐weighted means and variances of traits using gap‐filled data from the global plant trait database TRY. Moreover, we created a phylogenetic tree for 50,167 out of the 54,519 species identified in the plots. We present the first maps of global patterns of community richness and community‐weighted means of key traits. Conclusions: The availability of vegetation plot data in sPlot offers new avenues for vegetation analysis at the global scale.
Background Identifying common factors that affect public adherence to COVID-19 containment measures can directly inform the development of official public health communication strategies. The present international longitudinal study aimed to examine whether prosociality, together with other theoretically derived motivating factors (self-efficacy, perceived susceptibility and severity of COVID-19, perceived social support) predict the change in adherence to COVID-19 containment strategies. Method In wave 1 of data collection, adults from eight geographical regions completed online surveys beginning in April 2020, and wave 2 began in June and ended in September 2020. Hypothesized predictors included prosociality, self-efficacy in following COVID-19 containment measures, perceived susceptibility to COVID-19, perceived severity of COVID-19 and perceived social support. Baseline covariates included age, sex, history of COVID-19 infection and geographical regions. Participants who reported adhering to specific containment measures, including physical distancing, avoidance of non-essential travel and hand hygiene, were classified as adherence. The dependent variable was the category of adherence, which was constructed based on changes in adherence across the survey period and included four categories: non-adherence, less adherence, greater adherence and sustained adherence (which was designated as the reference category). Results In total, 2189 adult participants (82% female, 57.2% aged 31–59 years) from East Asia (217 [9.7%]), West Asia (246 [11.2%]), North and South America (131 [6.0%]), Northern Europe (600 [27.4%]), Western Europe (322 [14.7%]), Southern Europe (433 [19.8%]), Eastern Europe (148 [6.8%]) and other regions (96 [4.4%]) were analyzed. Adjusted multinomial logistic regression analyses showed that prosociality, self-efficacy, perceived susceptibility and severity of COVID-19 were significant factors affecting adherence. Participants with greater self-efficacy at wave 1 were less likely to become non-adherence at wave 2 by 26% (adjusted odds ratio [aOR], 0.74; 95% CI, 0.71 to 0.77; P < .001), while those with greater prosociality at wave 1 were less likely to become less adherence at wave 2 by 23% (aOR, 0.77; 95% CI, 0.75 to 0.79; P = .04). Conclusions This study provides evidence that in addition to emphasizing the potential severity of COVID-19 and the potential susceptibility to contact with the virus, fostering self-efficacy in following containment strategies and prosociality appears to be a viable public health education or communication strategy to combat COVID-19.
This study aimed to compare the mediation of psychological flexibility, prosociality and coping in the impacts of illness perceptions toward COVID-19 on mental health among seven regions. Convenience sampled online survey was conducted between April and June 2020 from 9130 citizens in 21 countries. Illness perceptions toward COVID-19, psychological flexibility, prosociality, coping and mental health, socio-demographics, lockdown-related variables and COVID-19 status were assessed. Results showed that psychological flexibility was the only significant mediator in the relationship between illness perceptions toward COVID-19 and mental health across all regions (all ps = 0.001–0.021). Seeking social support was the significant mediator across subgroups (all ps range = <0.001–0.005) except from the Hong Kong sample (p = 0.06) and the North and South American sample (p = 0.53). No mediation was found for problem-solving (except from the Northern European sample, p = 0.009). Prosociality was the significant mediator in the Hong Kong sample (p = 0.016) and the Eastern European sample (p = 0.008). These findings indicate that fostering psychological flexibility may help to mitigate the adverse mental impacts of COVID-19 across regions. Roles of seeking social support, problem-solving and prosociality vary across regions.
Background The COVID-19 pandemic triggered vast governmental lockdowns. The impact of these lockdowns on mental health is inadequately understood. On the one hand such drastic changes in daily routines could be detrimental to mental health. On the other hand, it might not be experienced negatively, especially because the entire population was affected. Methods The aim of this study was to determine mental health outcomes during pandemic induced lockdowns and to examine known predictors of mental health outcomes. We therefore surveyed n = 9,565 people from 78 countries and 18 languages. Outcomes assessed were stress, depression, affect, and wellbeing. Predictors included country, sociodemographic factors, lockdown characteristics, social factors, and psychological factors. Results Results indicated that on average about 10% of the sample was languishing from low levels of mental health and about 50% had only moderate mental health. Importantly, three consistent predictors of mental health emerged: social support, education level, and psychologically flexible (vs. rigid) responding. Poorer outcomes were most strongly predicted by a worsening of finances and not having access to basic supplies. Conclusions These results suggest that on whole, respondents were moderately mentally healthy at the time of a population-wide lockdown. The highest level of mental health difficulties were found in approximately 10% of the population. Findings suggest that public health initiatives should target people without social support and those whose finances worsen as a result of the lockdown. Interventions that promote psychological flexibility may mitigate the impact of the pandemic.
The coronavirus disease (COVID-19) pandemic fundamentally disrupted humans’ social life and behavior. Public health measures may have inadvertently impacted how people care for each other. This study investigated prosocial behavior, its association well-being, and predictors of prosocial behavior during the first COVID-19 pandemic lockdown and sought to understand whether region-specific differences exist. Participants (N = 9,496) from eight regions clustering multiple countries around the world responded to a cross-sectional online-survey investigating the psychological consequences of the first upsurge of lockdowns in spring 2020. Prosocial behavior was reported to occur frequently. Multiple regression analyses showed that prosocial behavior was associated with better well-being consistently across regions. With regard to predictors of prosocial behavior, high levels of perceived social support were most strongly associated with prosocial behavior, followed by high levels of perceived stress, positive affect and psychological flexibility. Sociodemographic and psychosocial predictors of prosocial behavior were similar across regions.
Background: The COVID-19 pandemic is a massive health crisis that has exerted enormous physical and psychological pressure. Mental healthcare for healthcare workers (HCWs) should receive serious consideration. This study served to determine the mental-health outcomes of 1,556 HCWs from 45 countries who participated in the COVID-19 IMPACT project, and to examine the predictors of the outcomes during the first pandemic wave. Methods: Outcomes assessed were self-reported perceived stress, depression symptom, and sleep changes. The predictors examined included sociodemographic factors and perceived social support. Results: The results demonstrated that half of the HCWs had moderate levels of perceived stress and symptoms of depression. Half of the HCWs (n = 800, 51.4%) had similar sleeping patterns since the pandemic started, and one in four slept more or slept less. HCWs reported less perceived stress and depression symptoms and higher levels of perceived social support than the general population who participated in the same project. Predictors associated with higher perceived stress and symptoms of depression among HCWs included female sex, not having children, living with parents, lower educational level, and lower social support. Discussion: The need for establishing ways to mitigate mental-health risks and adjusting psychological interventions and support for HCWs seems to be significant as the pandemic continues.
Abstract Plant functional traits directly affect ecosystem functions. At the species level, trait combinations depend on trade-offs representing different ecological strategies, but at the community level trait combinations are expected to be decoupled from these trade-offs because different strategies can facilitate co-existence within communities. A key question is to what extent community-level trait composition is globally filtered and how well it is related to global versus local environmental drivers. Here, we perform a global, plot-level analysis of trait–environment relationships, using a database with more than 1.1 million vegetation plots and 26,632 plant species with trait information. Although we found a strong filtering of 17 functional traits, similar climate and soil conditions support communities differing greatly in mean trait values. The two main community trait axes that capture half of the global trait variation (plant stature and resource acquisitiveness) reflect the trade-offs at the species level but are weakly associated with climate and soil conditions at the global scale. Similarly, within-plot trait variation does not vary systematically with macro-environment. Our results indicate that, at fine spatial grain, macro-environmental drivers are much less important for functional trait composition than has been assumed from floristic analyses restricted to co-occurrence in large grid cells. Instead, trait combinations seem to be predominantly filtered by local-scale factors such as disturbance, fine-scale soil conditions, niche partitioning and biotic interactions.
A population-based cross-sectional study was conducted during the first COVID-19 wave, to examine the impact of COVID-19 on mental health using an anonymous online survey, enrolling 9565 individuals in 78 countries. The current sub-study examined the impact of the pandemic and the associated lockdown measures on the mental health, and protective behaviors of cancer patients in comparison to non-cancer participants. Furthermore, 264 participants from 30 different countries reported being cancer patients. The median age was 51.5 years, 79.9% were female, and 28% had breast cancer. Cancer participants reported higher self-efficacy to follow recommended national guidelines regarding COVID-19 protective behaviors compared to non-cancer participants (p < 0.01). They were less stressed (p < 0.01), more psychologically flexible (p < 0.01), and had higher levels of positive affect compared to non-cancer participants. Amongst cancer participants, the majority (80.3%) reported COVID-19, not their cancer, as their priority during the first wave of the pandemic and females reported higher levels of stress compared to males. In conclusion, cancer participants appeared to have handled the unpredictable nature of the first wave of the pandemic efficiently, with a positive attitude towards an unknown and otherwise frightening situation. Larger, cancer population specific and longitudinal studies are warranted to ensure adequate medical and psychological care for cancer patients.
Abstract Parkinson’s disease (PD), with its characteristic loss of nigrostriatal dopaminergic neurons and deposition of α-synuclein in neurons, is often considered a neuronal disorder. However, in recent years substantial evidence has emerged to implicate glial cell types, such as astrocytes and microglia. In this study, we used stratified LD score regression and expression-weighted cell-type enrichment together with several brain-related and cell-type-specific genomic annotations to connect human genomic PD findings to specific brain cell types. We found that PD heritability attributable to common variation does not enrich in global and regional brain annotations or brain-related cell-type-specific annotations. Likewise, we found no enrichment of PD susceptibility genes in brain-related cell types. In contrast, we demonstrated a significant enrichment of PD heritability in a curated lysosomal gene set highly expressed in astrocytic, microglial, and oligodendrocyte subtypes, and in LoF-intolerant genes, which were found highly expressed in almost all tested cellular subtypes. Our results suggest that PD risk loci do not lie in specific cell types or individual brain regions, but rather in global cellular processes detectable across several cell types.