Equivalent analysis model of a GaN LED used as a receiver.

In general, visible light communication (VLC) uses LEDs as transmitters. However, LEDs can serve as receivers to construct a simple duplex VLC system that uses only two LEDs instead of one LED and one photo-diode (PD). There is a lack of effective equivalent analysis models for characterizing and evaluating the inherent behavioral characteristics of LEDs used as receivers. This paper presents an equivalent analysis model for GaN LEDs as receivers. First, based on the proposed receiving equivalent circuit model, a third-order signal transmission mathematical analysis model is established, revealing the transmission relationship between the photocurrent and output voltage. Further research is conducted on the impact of parameter changes on the bandwidth, and the model can be simplified into a first-order low-pass mathematical analysis model under specific conditions, providing theoretical support for improving the bandwidth of LED receiving applications. The experimental results also confirm the theoretical predictions. This research result holds significant importance for revealing the intrinsic mechanisms and the improved optical communication performance of LEDs for effective reception.

Advancing robust underwater acoustic target recognition through multitask learning and multi-gate mixture of experts.

Underwater acoustic target recognition has emerged as a prominent research area within the field of underwater acoustics. However, the current availability of authentic underwater acoustic signal recordings remains limited, which hinders data-driven acoustic recognition models from learning robust patterns of targets from a limited set of intricate underwater signals, thereby compromising their stability in practical applications. To overcome these limitations, this study proposes a recognition framework called M3 (multitask, multi-gate, multi-expert) to enhance the model's ability to capture robust patterns by making it aware of the inherent properties of targets. In this framework, an auxiliary task that focuses on target properties, such as estimating target size, is designed. The auxiliary task then shares parameters with the recognition task to realize multitask learning. This paradigm allows the model to concentrate on shared information across tasks and identify robust patterns of targets in a regularized manner, thus, enhancing the model's generalization ability. Moreover, M3 incorporates multi-expert and multi-gate mechanisms, allowing for the allocation of distinct parameter spaces to various underwater signals. This enables the model to process intricate signal patterns in a fine-grained and differentiated manner. To evaluate the effectiveness of M3, extensive experiments were implemented on the ShipsEar underwater ship-radiated noise dataset. The results substantiate that M3 has the ability to outperform the most advanced single-task recognition models, thereby achieving the state-of-the-art performance.

Adversarial multi-task underwater acoustic target recognition: Toward robustness against various influential factors.

Underwater acoustic target recognition based on passive sonar faces numerous challenges in practical maritime applications. One of the main challenges lies in the susceptibility of signal characteristics to diverse environmental conditions and data acquisition configurations, which can lead to instability in recognition systems. While significant efforts have been dedicated to addressing these influential factors in other domains of underwater acoustics, they are often neglected in the field of underwater acoustic target recognition. To overcome this limitation, this study designs auxiliary tasks that model influential factors (e.g., source range, water column depth, or wind speed) based on available annotations and adopts a multi-task framework to connect these factors to the recognition task. Furthermore, we integrate an adversarial learning mechanism into the multi-task framework to prompt the model to extract representations that are robust against influential factors. Through extensive experiments and analyses on the ShipsEar dataset, our proposed adversarial multi-task model demonstrates its capacity to effectively model the influential factors and achieve state-of-the-art performance on the 12-class recognition task.

Percutaneous endoscopic lumbar discectomy for single and double segment lumbar disc herniation with sciatic scoliosis in adults: a retrospective study.

OBJECTIVE: Sciatic scoliosis can be seen in patients with lumbar disc herniation. Percutaneous endoscopic lumbar discectomy (PELD) is a common surgical method for the treatment of lumbar disc herniation. The difference between single-segment lumbar disc herniation and double-segment lumbar disc herniation with Sciatic Scoliosis in adults after PELD needs further study. The aim of this study was to compare the imaging features of single-segment and double-segment lumbar disc herniation with Sciatic Scoliosis in adults and to further explore the clinical outcomes of functional improvement and scoliosis imaging parameters of the two groups after PELD. METHODS: Adult patients with lumbar disc herniation with sciatic scoliosis who received PELD from January 2019 to June 2022 were analyzed retrospectively. According to the number of operative segments, the patients were divided into a single-segment group and a double-segment group. Perioperative parameters were observed and compared between the two groups. The Visual Analogue Scale (VAS) score, Oswestry dysfunction index (ODI), Japanese Orthopaedic Association scores (JOA) and imaging parameters of the two groups were recorded and compared before the operation and during the follow-up. RESULTS: A total of 53 patients with single segments and 21 patients with double segments were included in this study. During the follow-up, the VAS score, ODI index and JOA score of the two groups were significantly improved as compared with those before the operation(P < 0. 05). Ninety-two point five percent of single-segment patients and 90.5% of double segment patients returned to normal scoliosis within 12 months after the operation. The operation time, number of intraoperative fluoroscopy times and the amount of intraoperative blood loss in single-segment patients were better than those in double-segment group(P < 0. 05). At the last follow-up, the AVT, CBD and SVA in the double-segment group were 5.2 ± 2.3, 5.1 ± 1.0 and 12.2 ± 3.0 mm, respectively, which were higher than those in the single-segment group (1.9 ± 0.4, 1.1 ± 1.6 and 3.9 ± 2.1 mm) (P < 0. 05). CONCLUSION: PELD is an effective treatment for single-segment and double-segment lumbar disc herniation with Sciatic scoliosis. Double-segment patients can enjoy similar clinical efficacy to single-segment patients, avoiding complications caused by decompression, fusion, and internal fixation. Scoliosis was corrected spontaneously within 12 months after operation, and the sagittal curve was significantly improved in both groups. The improvement of coronal and sagittal balance in double -segment patients may take longer.

Underwater-art: Expanding information perspectives with text templates for underwater acoustic target recognition.

Underwater acoustic target recognition is an intractable task due to the complex acoustic source characteristics and sound propagation patterns. Limited by insufficient data and narrow information perspective, recognition models based on deep learning seem far from satisfactory in practical underwater scenarios. Although underwater acoustic signals are severely influenced by distance, channel depth, or other factors, annotations of relevant information are often nonuniform, incomplete, and hard to use. In this work, the proposal is to implement underwater acoustic recognition based on templates made up of rich relevant information (UART). The templates are designed to integrate relevant information from different perspectives into descriptive natural language. UART adopts an audio-spectrogram-text trimodal contrastive learning framework, which endows UART with the ability to guide the learning of acoustic representations by descriptive natural language. These experiments reveal that UART has better recognition capability and generalization performance than traditional paradigms. Furthermore, the pretrained UART model could provide superior prior knowledge for the recognition model in the scenario without any auxiliary annotation.

Recognition efficiency of atypical cardiovascular readings on ECG devices through fogged goggles.

In their continuing battle against the COVID-19 pandemic, medical workers in hospitals worldwide need to wear safety glasses and goggles to protect their eyes from the possible transmission of the virus. However, they work for long hours and need to wear a mask and other personal protective equipment, which causes their protective eye wear to fog up. This fogging up of eye wear, in turn, has a substantial impact in the speed and accuracy of reading information on the interface of electrocardiogram (ECG) machines. To gain a better understanding of the extent of the impact, this study experimentally simulates the fogging of protective goggles when viewing the interface with three variables: the degree of fogging of the goggles, brightness of the screen, and color of the font of the cardiovascular readings. This experimental study on the target recognition of digital font is carried out by simulating the interface of an ECG machine and readability of the ECG machine with fogged eye wear. The experimental results indicate that the fogging of the lenses has a significant impact on the recognition speed and the degree of fogging has a significant correlation with the font color and brightness of the screen. With a reduction in screen brightness, its influence on recognition speed shows a v-shaped trend, and the response time is the shortest when the screen brightness is 150 cd/m2. When eyewear is fogged, yellow and green font colors allow a quicker response with a higher accuracy. On the whole, the subjects show a better performance with the use of green font, but there are inconsistencies. In terms of the interaction among the three variables, the same results are also found and the same conclusion can be made accordingly. This research study can act as a reference for the interface design of medical equipment in events where medical staff wear protective eyewear for a long period of time.

The profile of gut microbiota and central carbon-related metabolites in primary angle-closure glaucoma patients.

PURPOSE: To explore the profile of gut microbiota and central carbon-related metabolites in patients with primary angle-closure glaucoma (PACG). METHODS: The fecal microbiotas of 30 PACG patients and 30 healthy participants were detected via 16S rRNA sequencing. Targeted liquid chromatography-mass spectrometry was used to examine serum central carbon-related metabolites. The correlations among metabolites, microbiotas and clinical presentations were also explored. RESULTS: Although the α and ß diversity between the PACG and control groups did not show a significant difference, the distribution of Blautia and Fusicatenibacter decreased significantly in the PACG group. Functional annotations of microbiota enrichment showed that the most dominant pathway was related to host metabolism. In the PACG patients, seven central carbon metabolites, namely adenosine 5'-diphosphate, dGDP, phosphoenolpyruvic acid, d-ribulose 5-phosphate, d-xylulose 5-phosphate, glucuronic acid, and malonic acid, decreased significantly, whereas two metabolites, citric acid and isocitrate, increased obviously. The mean RNFL thickness was positively correlated with phosphoenolpyruvic acid, the VF-MD was positively correlated with glucuronic acid, and the abundance of Blautia was negatively associated with citric acid. CONCLUSION: Few species of gut microbiota were altered in the PACG patients compared to the healthy subjects. A distinct difference in the phenotype of the central carbon-related metabolites of PACG and their correlation with clinical presentations and microbiota suggests potential mechanisms of RGC impairment and novel intervention targets.

A Feature Optimization Approach Based on Inter-Class and Intra-Class Distance for Ship Type Classification.

Deep learning based methods have achieved state-of-the-art results on the task of ship type classification. However, most existing ship type classification algorithms take time-frequency (TF) features as input, the underlying discriminative information of these features has not been explored thoroughly. This paper proposes a novel feature optimization method which is designed to minimize an objective function aimed at increasing inter-class and reducing intra-class feature distance for ship type classification. The objective function we design is able to learn a center for each class and make samples from the same class closer to the corresponding center. This ensures that the features maximize underlying discriminative information involved in the data, particularly for some targets that usually confused by the conventional manual designed feature. Results on the dataset from a real environment show that the proposed feature optimization approach outperforms traditional TF features.

Volatile fatty acids and biogas recovery using thermophilic anaerobic membrane distillation bioreactor for wastewater reclamation.

The effects of bioreactor temperatures and salinities of an anaerobic membrane distillation bioreactor (anMDBR) on the permeation performance and their potential recovery of bioresources were fully examined in this study. To the best of our knowledge, this is the first study of a lab-scale anMDBR process utilizing sub-merged hollow fiber membranes. The hybrid system utilizing both membrane distillation (MD) and anaerobic bioreactors achieved 99.99% inorganic salt rejection regardless the operation temperatures and high initial flux from (2-4 L m-2 h-1) at 45-65 °C. However, after 7-day operation, the flux dropped by 16-50% proportional to the bioreactor temperatures. It was found that the effects of bioreactor temperatures had strong impacts on both the permeation performance and fouling behavior while salinity had insignificant effect. A compact non-porous fouling layer was observed on the membrane surface from the bioreactor operated at 65 °C while only a few depositions was found on the membrane from 45 °C bioreactor. In the present study, the optimal anMDBR temperature was found to be 45 °C, showing a balanced biogas production and membrane permeation performance including less fouling formation. At this bioreactor temperature (45 °C), the biogas yield was 0.14 L/g CODremoval, while maintaining a methane recovery of 42% in the biogas, similar recovery to those at bioreactor temperatures of 55 and 65 °C. The potential recovery of volatile fatty acids made anMDBR a more economically efficient system, in addition to its lower operation cost and smaller footprint compared with most other technologies for on-site wastewater treatment.

Noninvasive Multimodal Imaging of Osteosarcoma and Lymph Nodes Using a 99mTc-Labeled Biomineralization Nanoprobe.

The accurate imaging of the lymph nodes represents a critical indicator for tumor staging and surgical planning (e.g., osteosarcoma). Clinically, nodal tracing using a radio-nanocolloid is often limited by the inaccessibility of real-time images and inadequate anatomical information. Herein, we present a 99mTc-labeled biomineralization nanoprobe for the advanced detection of osteosarcoma and lymph nodes with multimodal imaging. Through the exploitation of the complementary strengths of MRI/SPECT/NIR fluorescence, the fabricated nanoprobe exhibited suitable stability and biocompatibility characteristics and was shown to be able to be located in osteosarcoma. The lymphatic drainage and network in healthy mice were imaged in real-time using NIR fluorescence and SPECT/CT. Furthermore, we demonstrated that our 99mTc-biomineralization nanoprobe could be used for the high-resolution and high-sensitivity imaging analysis of lymphatic drainage in an orthotopic osteosarcoma model. Overall, the 99mTc-labeled biomineralization nanoprobe features promising characteristics to be used as an intraoperative visualization tool to aid in precise tumor imaging and nodal resection.

Epithelial phenotype as a predictive marker for response to EGFR-TKIs in non-small cell lung cancer patients with wild-type EGFR.

Epithelial-to-mesenchymal transition (EMT) has profound impacts on cancer progression and also on drug resistance, including epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs). Nowadays, there is still no predictive biomarker identified for the use of EGFR-TKIs in non-small cell lung cancer (NSCLC) patients with wild-type EGFR. To clarify the role of EMT phenotype as a predictive marker for EGFR-TKI, we performed a retrospective study in 202 stage IV or recurrent NSCLC patients receiving gefitinib or erlotinib therapy from June 2008 to September 2012 in our institute. Clinical data and EGFR mutational status were collected, while epithelial, epithelial to mesenchymal, not specified or mesenchymal phenotype were classified according to EMT markers such as E-cadherin, fibronectin, N-cadherin and vimentin by immunohistochemistry. Epithelial phenotype was more frequently found in patients with EGFR mutation (p = 0.044). Epithelial phenotype was associated with a significantly higher objective response rate (23.5 vs. 11.1 vs. 0.0 vs. 2.4%, p = 0.011), longer progression-free survival (4.4 vs. 1.9 vs. 1.7 vs. 1.0 months, p < 0.001) and longer overall survival (11.5 vs. 8.9 vs. 4.5 vs. 4.9 months, p < 0.001) compared to epithelial to mesenchymal, not specified and mesenchymal phenotype in the wild-type EGFR subgroup. In the subgroup with EGFR mutation, the trend remained but without a statistically significant difference. In conclusion, epithelial phenotype was more likely expressed in patients with EGFR mutation and was associated with a better outcome in advanced NSCLC patients with wild-type EGFR, which indicates that the EMT phenotype might be a potential marker to guide EGFR-TKI therapy in this population.

Promoting positive youth development in rural communities: Integrating social work, psychology, and education.

Considering the peculiar socio-cultural background and developmental obstacles encountered by rural youth in China, the study examines the necessity of adopting an integrated strategy that brings together social work, psychology, and education to promote positive youth development. This research intends to fill the gap by explaining the impact of these factors on community engagement and youth development in China. Targeted programs were also suggested according to the needs of rural youth in China. The respondents of the study comprised 350 young people, whose age ranged from 15 to 24 years, living in different rural areas of the country. The structured questionnaire was designed to collect the data using a convenience sampling technique. Structural Equation Modeling (SEM) was applied as the analysis tool using IBM SPSS AMOS software. The results show that social work and education have a significant impact on community engagement and positive youth development. The findings also reveal that psychology positively influences community engagement. Community engagement was seen to mediate the relationships between social work, psychology, education, and positive youth development. The policymakers and practitioners can fully use the interrelationships between social work, psychology, and education to create a more comprehensive approach that considers the specific characteristics of rural youth in China. Additionally, highlighting community engagement as a mediator also explores the opportunity for bottom-up initiatives and community efforts to instigate favorable youth outcomes in the countryside.

Scalable Hierarchically Structured Materials from a Multiscale Particle System Enabled by Microscaffolds.

Structural hierarchy is the key to manufacturing multiscale particle-based composite materials. A novel manufacturing method was developed to generate scalable hierarchical structures in concrete. The new method used 3D-printed microscaffolds to interact with the multiscale particle packing in concrete, resulting in a structured lightweight composite material. The size of internal members can vary by more than two orders of magnitude, to adapt to different applications. Based on compression tests and microstructural investigation by optical microscope and quantitative nanomechanical mapping, we found that the new material is 63.93% more efficient in energy absorption capacity compared with traditional lightweight concrete. Our experimental trials also showed that introducing structural hierarchy can reduce the consumption of cementitious material in the system by up to 14% and significantly reduce the use of scaffolds. The method could be applied to a board spectrum of multiscale particle-based materials, such as dental cement and bone implant materials, to improve material performance and efficiency in medical and construction applications.

Impacts of sulfamethoxazole stress on vegetable growth and rhizosphere bacteria and the corresponding mitigation mechanism.

Antibiotics are an important pharmaceutical class excessively used by humans. Its presence in the soil can impact plant growth and induce antibiotic resistance. This research studies the effect of sulfamethoxazole (SMX) on plant growth, rhizosphere bacteria composition, and resistance genes. Two sets of vegetables (basil, cilantro, and spinach) were treated separately with water and SMX solution. The plant growth data and soil samples were collected and analyzed. The results revealed that SMX increased spinach leaf length (34.0%) while having no significant impacts on basil and cilantro. On the other hand, SMX improved the bacterial diversity in all samples. The shifts in the abundance of plant growth-promoting bacteria could indirectly affect vegetable stem and leaf length. SMX also significantly increased the abundance of resistance genes Sul1 and Sul2. A further study into the correlation between bacteria highlights the importance of Shingomonas and Alfipia for inhibiting the spread of key resistance gene hosts, namely, Pseudomonas, Stenotrophomonas, and Agrobacterium. This research provides insight into SMX's impact on vegetable growth and microbial diversity. It also points out important microbial interactions that could potentially be utilized to mitigate ARG proliferation.

Recent Advances in High-Rate Solar-Driven Interfacial Evaporation.

Recent advances in solar-driven interfacial evaporation (SDIE) have led to high evaporation rates that open promising avenues for practical utilization in freshwater production and industrial application for pollutant and nutrient concentration, and resource recovery. Breakthroughs in overcoming the theoretical limitation of 2D interfacial evaporation have allowed for developing systems with high evaporation rates. This study presents a comprehensive review of various evaporator designs that have achieved pure evaporation rates beyond 4 kg m-2 h-1, including structural and material designs allowing for rapid evaporation, passive 3D designs, and systems coupled with alternative energy sources of wind and joule heating. The operational mechanisms for each design are outlined together with discussion on the current benefits and areas for improvement. The overarching challenges encountered by SDIE concerning the feasibility of direct integration into contemporary practical settings are assessed, and issues relating to sustaining elevated evaporation rates under diverse environmental conditions are addressed.

Reliable assessment and prediction of moderate preoxidation of sodium hypochlorite for algae-laden water treatment.

Chemical moderate preoxidation for algae-laden water is an economical and prospective strategy for controlling algae and exogenous pollutants, whereas it is constrained by a lack of effective on-line evaluation and quick-response feedback method. Herein, excitation-emission matrix parallel factor analysis (EEM-PARAFAC) was used to identify cyanobacteria fluorophores after preoxidation of sodium hypochlorite (NaClO) at Excitation/Emission wavelength of 260(360)/450 nm, based on which the algal cell integrity and intracellular organic matter (IOM) release were quantitatively assessed. Machine learning modeling of fluorescence spectral data for prediction of moderate preoxidation using NaClO was established. The optimal NaClO dosage for moderate preoxidation depended on algal density, growth phases, and organic matter concentrations in source water matrices. Low doses of NaClO (<0.5 mg/L) led to short-term desorption of surface-adsorbed organic matter (S-AOM) without compromising algal cell integrity, whereas high doses of NaClO (≥0.5 mg/L) quickly caused cell damage. The optimal NaClO dosage increased from 0.2-0.3 mg/L to 0.9-1.2 mg/L, corresponding to the source water with algal densities from 0.1 × 106 to 2.0 × 106 cells/mL. Different growth stages required varying NaClO doses: stationary phase cells needed 0.3-0.5 mg/L, log phase cells 0.6-0.8 mg/L, and decaying cells 2.0-2.5 mg/L. The presence of natural organic matter and S-AOM increased the NaClO dosage limit with higher dissolved organic carbon (DOC) concentrations (1.00 mg/L DOC required 0.8-1.0 mg/L NaClO, while 2.20 mg/L DOC required 1.5-2.0 mg/L). Compared to other predictive models, the machine learning model (Gaussian process regression-Matern (0.5)) performed best, achieving R2 values of 1.000 and 0.976 in training and testing sets. Optimal preoxidation followed by coagulation effectively removed algal contaminants, achieving 91%, 92%, and 92% removal for algal cells, turbidity, and chlorophyll-a, respectively, thereby demonstrating the effectiveness of moderate preoxidation. This study introduces a novel approach to dynamically adjust NaClO dosage by monitoring source water qualities and tracking post-preoxidation fluorophores, enhancing moderate preoxidation technology application in algae-laden water treatment.

The Burden and Clinical Features of Neovascular Glaucoma in A Major Tertiary Care Center in China.

PRCIS: This research presents the burden and clinical characteristics of NVG in Zhongshan Ophthalmic Center, employing the most extensive sample size and the longest uninterrupted temporal duration so far, which may provide a theoretical reference for the effective prevention and diagnosis of NVG. PURPOSE: To summarize the burden and clinical characteristics of neovascular glaucoma (NVG) in a major tertiary care center in China. METHODS: The clinical data of NVG patients in Zhongshan Ophthalmic Center (ZOC) between 2012 and 2021 were collected retrospectively, including their age, sex, affected eye, best-corrected visual acuity (BCVA), intraocular pressure (IOP), clinical stage and aetiology. RESULTS: In this study, we included 1877 eyes of 1749 patients who developed NVG, with 2.01:1 ratio of male to female. Their mean age was 53.14±16.69 years and those aged 41-70 years (65.2%) were most affected. Monocular patients were more predominant in most of them (92.7%), while 7.3% were binocular and 1667 eyes (88.8%) were at the angle­closure stage. The BCVA and IOP were 2.42±0.70 logMAR and 38.6±12.2 mmHg, respectively. Over the decade, the number of NVG patients and the proportion of NVG patients among glaucoma patients showed an increasing trend, with annual percentage changes (APCs) of 9.1% (95% CI: 5.0-13.3%, P=0.001) and 4.8% (95% CI: 2.2-7.4%, P=0.003), respectively. The top three primary conditions were diabetic retinopathy (DR), retinal vein occlusion (RVO), and retinal detachment (RD). Moreover, the APCs for the constituent ratio of DR and RVO were 4.4% (95% CI: 0.5-8.4%, P=0.031) and ï¹£4.6% (95% CI: ï¹£8.4% to ï¹£0.7%, P=0.028), respectively. However, the first and second causes of NVG in minors (<18 years old) were Coat's disease and ocular tumours, followed by RD and RVO in third place. The top cause of NVG in patients aged 65 years and older was RVO. CONCLUSIONS: The burden of NVG is increasing, emphasizing the need to improve preventive strategies focusing on primary diseases such as DR, RVO, and RD, particularly the increasing proportion of DR cases and the previously underemphasized RD patients, while also highlighting the differences in primary diseases across different age groups.

Hydrodynamic solar-driven interfacial evaporation - Gone with the flow.

Evaporation has been one of the most classic desalination processes on the Earth. When we try to use the power of water flow itself, the evaporation process can perform even better. Here, we report a hydrodynamic solar-driven interfacial evaporation process which water evaporation rate can achieve 6.58 kg·m-2·h-1 (over 100 times higher than natural evaporation). A waterwheel-structure solar interfacial evaporator was designed and assembled by printed filter papers. The evaporator can both rapidly distribute solution on the evaporation interface and be hydraulically driven to rotate continuously to improve the evaporation rate by water flow. The hydrodynamic solar-driven interfacial evaporation process successfully overcomes the problem of slow diffusion of water vapor, but also realizes the day-and-night operation of process and the self-cleaning of salt fouling. Apart from the application in solar desalination, the developed evaporator has great potentials in vapor production and salt recovery for industrial use.

Stress Suppression Design for Radiofrequency Microelectromechanical System Switch Based on a Flexible Substrate.

A novel stress suppression design for flexible RF MEMS switches has been presented and demonstrated through theoretical and experimental research to isolate the stress caused by substrate bending. An RF MEMS switch with an S-shaped microspring structure was fabricated by the two-step etching process as a developmental step toward miniaturization and high reliability. The RF MEMS switches with an S-shaped microspring exhibited superior microwave performance and stable driving voltage under different substrate curvatures compared to the conventional non-microspring switches, demonstrating that the bending stress is successfully suppressed by the S-shaped microspring and the island structure. Furthermore, this innovative design could be easily extended to other flexible devices.

Playing for 3D Human Recovery.

Image- and video-based 3D human recovery (i.e., pose and shape estimation) have achieved substantial progress. However, due to the prohibitive cost of motion capture, existing datasets are often limited in scale and diversity. In this work, we obtain massive human sequences by playing the video game with automatically annotated 3D ground truths. Specifically, we contribute GTA-Human, a large-scale 3D human dataset generated with the GTA-V game engine, featuring a highly diverse set of subjects, actions, and scenarios. More importantly, we study the use of game-playing data and obtain five major insights. First, game-playing data is surprisingly effective. A simple frame-based baseline trained on GTA-Human outperforms more sophisticated methods by a large margin. For videobased methods, GTA-Human is even on par with the in-domain training set. Second, we discover that synthetic data provides critical complements to the real data that is typically collected indoor. We highlight that our investigation into domain gap provides explanations for our data mixture strategies that are simple yet useful, which offers new insights to the research community. Third, the scale of the dataset matters. The performance boost is closely related to the additional data available. A systematic study on multiple key factors (such as camera angle and body pose) reveals that the model performance is sensitive to data density. Fourth, the effectiveness of GTA-Human is also attributed to the rich collection of strong supervision labels (SMPL parameters), which are otherwise expensive to acquire in real datasets. Fifth, the benefits of synthetic data extend to larger models such as deeper convolutional neural networks (CNNs) and Transformers, for which a significant impact is also observed. We hope our work could pave the way for scaling up 3D human recovery to the real world. Homepage: https://caizhongang.github.io/projects/GTA-Human/.