We propose Mixed-Precision Multibranch Network(M+MNet)to compensate for the neglect of background information in image aesthetics assessment(IAA)while providing strategies for overcoming the dilemma between training c...We propose Mixed-Precision Multibranch Network(M+MNet)to compensate for the neglect of background information in image aesthetics assessment(IAA)while providing strategies for overcoming the dilemma between training costs and performance.First,two exponentially weighted pooling methods are used to selectively boost the extraction of background and salient information during downsampling.Second,we propose Corner Grid,an unsupervised data augmentation method that leverages the diffusive characteristics of convolution to force the network to seek more relevant background information.Third,we perform mixed-precision training by switching the precision format,thus significantly reducing the time and memory consumption of data representation and transmission.Most of our methods specifically designed for IAA tasks have demonstrated generalizability to other IAA works.For performance verification,we develop a large-scale benchmark(the most comprehensive thus far)by comparing 17 methods with M+MNet on two representative datasets:the Aesthetic Visual Analysis(AVA)dataset and FLICKR-Aesthetic Evaluation Subset(FLICKR-AES).M+MNet achieves state-of-the-art performance on all tasks.展开更多
Developing chemical systems capable of executing multipath nonequilibrium processes provides valuable insights into the study of multistage dissipative structures in living systems.Here,we report a nonequilibrium mult...Developing chemical systems capable of executing multipath nonequilibrium processes provides valuable insights into the study of multistage dissipative structures in living systems.Here,we report a nonequilibrium multipath evolution in a three-state supramolecular system,which is clarified by associating it with a multibranch selection structure.This system is constructed by synergistically combining an assembly-induced sol-gel transition with the color change of an acid-base indicator.Depending on the initial condition and chemical fuel composition,this system can execute five distinct nonequilibrium processes.In addition,the set of structured programming can be replayed by repeatedly introducing the chemical fuel.It is further utilized to explore the applications of information encryption and establish its universality.These findings offer novel avenues for future transient material design and may make significant contributions to systems chemistry.展开更多
离线高精地图的生产,需要昂贵的经济和人力资源,因此,基于车载周视相机系统在线构建局部道路地图的方法,成为当前自动驾驶领域关注的焦点。但现有方案往往基于单帧数据,聚焦于较小范围内的道路地图构建性能,而忽略了自动驾驶场景中天然...离线高精地图的生产,需要昂贵的经济和人力资源,因此,基于车载周视相机系统在线构建局部道路地图的方法,成为当前自动驾驶领域关注的焦点。但现有方案往往基于单帧数据,聚焦于较小范围内的道路地图构建性能,而忽略了自动驾驶场景中天然存在的时序信息,及其对自车行进方向上更长范围建图的帮助。为此,本文提出一种融合场景级时空信息的长范围矢量道路地图构建算法。基于解耦矢量地图的语义属性和几何属性的思想,通过多分支特征编码机制,增强对地图元素空间位置信息的理解能力。融合多分支特征后,将当前帧的鸟瞰图(bird′s eye view,BEV)特征与场景时空信息进行关联,并结合语义分割作为辅助任务引导时空信息融合,有效应对行车过程中的遮挡问题。实验结果表明,在公开数据集nuScenes和Argoverse2上的长范围建图质量,分别取得了显著的18.7%和6.4%全类平均精度(mean average precision,MAP)提高。展开更多
A novel multi-branched ligand with multiple ester groups,multiple benzene rings and multiple carboxyls was synthesized.The composition and structure of the ligand were characterized through the Fourier transform infra...A novel multi-branched ligand with multiple ester groups,multiple benzene rings and multiple carboxyls was synthesized.The composition and structure of the ligand were characterized through the Fourier transform infrared spectroscopy(FT-IR),nuclear magnetic resonance spectrum(1H NMR) and UV spectrometer.The composition and fluorescence properties of rare earth complexes with the ligand and 1,10-phenanthroline(phen) were researched via FT-IR,EA and fluorescent spectrometer.The results presented that the resulting complex had strong characteristic fluorescence spectrum of the rare earth ions,which was attributed to the synergistic effects of the ligand,phen,and rare earth ion.展开更多
Pt-（Sn,Re）/HZSM5-HMS catalysts were evaluated for n-heptane isomerization at 200–350 ℃.To characterize the catalyst,X-ray diffraction,X-ray fluorescene,Fourier transform infrared spectroscopy,ultraviolet-visible d...Pt-（Sn,Re）/HZSM5-HMS catalysts were evaluated for n-heptane isomerization at 200–350 ℃.To characterize the catalyst,X-ray diffraction,X-ray fluorescene,Fourier transform infrared spectroscopy,ultraviolet-visible diffuse reflectance spectroscopy,temperature-programmed reduction of H2,temperature-programmed desorption of NH3,infrared spectroscopy of adsorbed pyridine,H2 chemisorption,nitrogen adsorption-desorption,scanning electron microscopy and thermogravimetric analysis were performed.Kinetics of n-C7 isomerization were investigated under various hydrogen and n-C7 pressures,and the effects of reaction conditions on catalytic performance were studied.The results showed that bi-and trimetallic catalysts exhibit better performance than monometallic catalysts for this reaction.For example,a maximum i-C7 selectivity（ 〉74%） and multibranched isomer selectivity（40%） were observed for Pt-Sn/HZSM5-HMS at 200 ℃.展开更多
We present a novel methodology and strategy to predict pressures and flow rates in the global cardiovascular network in different postures varying from supine to upright. A closed-loop, multiscale mathematical model o...We present a novel methodology and strategy to predict pressures and flow rates in the global cardiovascular network in different postures varying from supine to upright. A closed-loop, multiscale mathematical model of the entire cardiovascular system (CVS) is developed through an integration of one-dimensional (1D) modeling of the large systemic arteries and veins, and zero-dimensional (0D) lumped-parameter modeling of the heart, the cardiac-pulmonary circulation, the cardiac and venous valves, as well as the microcirculation. A versatile junction model is proposed and incorporated into the 1D model to cope with splitting and/or merging flows across a multibranched junction, which is validated to be capable of estimating both subcritical and supercritical flows while ensuring the mass conservation and total pressure continuity. To model gravitational effects on global hemodynamics during postural change, a robust venous valve model is further established for the 1D venous flows and distributed throughout the entire venous network with consideration of its anatomically realistic numbers and locations. The present integrated model is proven to enable reasonable prediction of pressure and flow rate waveforms associated with cardiopulmonary circulation, systemic circulation in arteries and veins, as well as microcirculation within normal physiological ranges, particularly in mean venous pressures, which well match the in vivo measurements. Applications of the cardiovascular model at different postures demonstrate that gravity exerts remarkable influence on arterial and venous pressures, venous returns and cardiac outputs whereas venous pressures below the heart level show a specific correlation between central venous and hydrostatic pressures in right atrium and veins.展开更多
We present a four-state model for calculating the two-photon absorption of multi-branched molecules by using the time-depended function method, The numerical results indicate that the two-photon absorption cross secti...We present a four-state model for calculating the two-photon absorption of multi-branched molecules by using the time-depended function method, The numerical results indicate that the two-photon absorption cross section has a strong enhancement for three-branch molecules compared to two-branch structures, The maximal two-photon-absorption cross section is 2.358 × 10^-47 cm^4s/photon, At the same time, the charge-transfer process for the charge-transfer states is visualized in order to explain mechanism about the maximal TPA cross section.展开更多
基金supported by the National Natural Science Foundation of China under Grant No.62502040the ZTE Industry-University-Institute Cooperation Funds under Grant No.IA20230700001.
文摘We propose Mixed-Precision Multibranch Network(M+MNet)to compensate for the neglect of background information in image aesthetics assessment(IAA)while providing strategies for overcoming the dilemma between training costs and performance.First,two exponentially weighted pooling methods are used to selectively boost the extraction of background and salient information during downsampling.Second,we propose Corner Grid,an unsupervised data augmentation method that leverages the diffusive characteristics of convolution to force the network to seek more relevant background information.Third,we perform mixed-precision training by switching the precision format,thus significantly reducing the time and memory consumption of data representation and transmission.Most of our methods specifically designed for IAA tasks have demonstrated generalizability to other IAA works.For performance verification,we develop a large-scale benchmark(the most comprehensive thus far)by comparing 17 methods with M+MNet on two representative datasets:the Aesthetic Visual Analysis(AVA)dataset and FLICKR-Aesthetic Evaluation Subset(FLICKR-AES).M+MNet achieves state-of-the-art performance on all tasks.
基金supported by the National Natural Science Foundation of China(grant no.21975145)the Natural Science Foundation of Shandong Province(grant no.ZR2023JQ008).
文摘Developing chemical systems capable of executing multipath nonequilibrium processes provides valuable insights into the study of multistage dissipative structures in living systems.Here,we report a nonequilibrium multipath evolution in a three-state supramolecular system,which is clarified by associating it with a multibranch selection structure.This system is constructed by synergistically combining an assembly-induced sol-gel transition with the color change of an acid-base indicator.Depending on the initial condition and chemical fuel composition,this system can execute five distinct nonequilibrium processes.In addition,the set of structured programming can be replayed by repeatedly introducing the chemical fuel.It is further utilized to explore the applications of information encryption and establish its universality.These findings offer novel avenues for future transient material design and may make significant contributions to systems chemistry.
文摘离线高精地图的生产,需要昂贵的经济和人力资源,因此,基于车载周视相机系统在线构建局部道路地图的方法,成为当前自动驾驶领域关注的焦点。但现有方案往往基于单帧数据,聚焦于较小范围内的道路地图构建性能,而忽略了自动驾驶场景中天然存在的时序信息,及其对自车行进方向上更长范围建图的帮助。为此,本文提出一种融合场景级时空信息的长范围矢量道路地图构建算法。基于解耦矢量地图的语义属性和几何属性的思想,通过多分支特征编码机制,增强对地图元素空间位置信息的理解能力。融合多分支特征后,将当前帧的鸟瞰图(bird′s eye view,BEV)特征与场景时空信息进行关联,并结合语义分割作为辅助任务引导时空信息融合,有效应对行车过程中的遮挡问题。实验结果表明,在公开数据集nuScenes和Argoverse2上的长范围建图质量,分别取得了显著的18.7%和6.4%全类平均精度(mean average precision,MAP)提高。
文摘目的 为了更好地实现轻量化的人体姿态估计,在轻量级模型极为有限的资源下实现更高的检测性能。基于高分辨率网络(high resolution network,HRNet)提出了结合密集连接网络的轻量级高分辨率人体姿态估计网络(lightweight high-resolution human estimation combined with densely connected network,LDHNet)。方法 通过重新设计HRNet中的阶段分支结构以及提出新的轻量级特征提取模块,构建了轻量高效的特征提取单元,同时对多分支之间特征融合部分进行了轻量化改进,进一步降低模型的复杂度,最终大幅降低了模型的参数量与计算量,实现了轻量化的设计目标,并且保证了模型的性能。结果 实验表明,在MPII(Max Planck Institute for Informatics)测试集上相比于自顶向下的轻量级人体姿态估计模型LiteHRNet,LDHNet仅通过增加少量参数量与计算量,平均预测准确度即提升了1.5%,与LiteHRNet的改进型DiteHRNet相比也提升了0.9%,在COCO(common objects in context)验证集上的结果表明,与LiteHRNet相比,LDHNet的平均检测准确度提升了3.4%,与DiteHRNet相比也提升了2.3%,与融合Transformer的HRFormer相比,LDHNet在参数量和计算量都更低的条件下有近似的检测性能,在面对实际场景时LDHNet也有着稳定的表现,在同样的环境下LDHNet的推理速度要高于基线HRNet以及LiteHRNet等。结论 该模型有效实现了轻量化并保证了预测性能。
基金Project supported by the Key Project of Chinese Ministry of Education (208005)the Young Scientist Backbone of Heilongjiang Common Higher University (1154G03)
文摘A novel multi-branched ligand with multiple ester groups,multiple benzene rings and multiple carboxyls was synthesized.The composition and structure of the ligand were characterized through the Fourier transform infrared spectroscopy(FT-IR),nuclear magnetic resonance spectrum(1H NMR) and UV spectrometer.The composition and fluorescence properties of rare earth complexes with the ligand and 1,10-phenanthroline(phen) were researched via FT-IR,EA and fluorescent spectrometer.The results presented that the resulting complex had strong characteristic fluorescence spectrum of the rare earth ions,which was attributed to the synergistic effects of the ligand,phen,and rare earth ion.
文摘Pt-（Sn,Re）/HZSM5-HMS catalysts were evaluated for n-heptane isomerization at 200–350 ℃.To characterize the catalyst,X-ray diffraction,X-ray fluorescene,Fourier transform infrared spectroscopy,ultraviolet-visible diffuse reflectance spectroscopy,temperature-programmed reduction of H2,temperature-programmed desorption of NH3,infrared spectroscopy of adsorbed pyridine,H2 chemisorption,nitrogen adsorption-desorption,scanning electron microscopy and thermogravimetric analysis were performed.Kinetics of n-C7 isomerization were investigated under various hydrogen and n-C7 pressures,and the effects of reaction conditions on catalytic performance were studied.The results showed that bi-and trimetallic catalysts exhibit better performance than monometallic catalysts for this reaction.For example,a maximum i-C7 selectivity（ 〉74%） and multibranched isomer selectivity（40%） were observed for Pt-Sn/HZSM5-HMS at 200 ℃.
基金supported by a Grant-in-Aid for Scientific Research (Grant 17300141)Japan Society for the Promotion of Science and Research and Development of the Next Generation Integrated Simulation of Living Matter, JST,a part of the Development and Use of the Next Generation Supercomputer Project of the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japanthe RIKEN Junior Research Associate Program
文摘We present a novel methodology and strategy to predict pressures and flow rates in the global cardiovascular network in different postures varying from supine to upright. A closed-loop, multiscale mathematical model of the entire cardiovascular system (CVS) is developed through an integration of one-dimensional (1D) modeling of the large systemic arteries and veins, and zero-dimensional (0D) lumped-parameter modeling of the heart, the cardiac-pulmonary circulation, the cardiac and venous valves, as well as the microcirculation. A versatile junction model is proposed and incorporated into the 1D model to cope with splitting and/or merging flows across a multibranched junction, which is validated to be capable of estimating both subcritical and supercritical flows while ensuring the mass conservation and total pressure continuity. To model gravitational effects on global hemodynamics during postural change, a robust venous valve model is further established for the 1D venous flows and distributed throughout the entire venous network with consideration of its anatomically realistic numbers and locations. The present integrated model is proven to enable reasonable prediction of pressure and flow rate waveforms associated with cardiopulmonary circulation, systemic circulation in arteries and veins, as well as microcirculation within normal physiological ranges, particularly in mean venous pressures, which well match the in vivo measurements. Applications of the cardiovascular model at different postures demonstrate that gravity exerts remarkable influence on arterial and venous pressures, venous returns and cardiac outputs whereas venous pressures below the heart level show a specific correlation between central venous and hydrostatic pressures in right atrium and veins.
基金Supported by the National Natural Science Foundation of China under Grant No 60471042.
文摘We present a four-state model for calculating the two-photon absorption of multi-branched molecules by using the time-depended function method, The numerical results indicate that the two-photon absorption cross section has a strong enhancement for three-branch molecules compared to two-branch structures, The maximal two-photon-absorption cross section is 2.358 × 10^-47 cm^4s/photon, At the same time, the charge-transfer process for the charge-transfer states is visualized in order to explain mechanism about the maximal TPA cross section.
