We consider a Prandtl model derived from MHD in the Prandtl-Hartmann regime that has a damping term due to the effect of the Hartmann boundary layer.A global-in-time well-posedness is obtained in the Gevrey function s...We consider a Prandtl model derived from MHD in the Prandtl-Hartmann regime that has a damping term due to the effect of the Hartmann boundary layer.A global-in-time well-posedness is obtained in the Gevrey function space with the optimal index 2.The proof is based on a cancellation mechanism through some auxiliary functions from the study of the Prandtl equation and an observation about the structure of the loss of one order tangential derivatives through twice operations of the Prandtl operator.展开更多
针对自动驾驶场景动态目标检测存在检测速度难以满足实时性要求、检测目标小或被遮挡造成的精度不足和误检、漏检率高等问题,提出一种基于改进YOLOv8模型的行人及车辆检测方法。首先,在Backbone骨干网络提取图像特征时使用对图像分辨率...针对自动驾驶场景动态目标检测存在检测速度难以满足实时性要求、检测目标小或被遮挡造成的精度不足和误检、漏检率高等问题,提出一种基于改进YOLOv8模型的行人及车辆检测方法。首先,在Backbone骨干网络提取图像特征时使用对图像分辨率低、小目标检测友好的空间到深度卷积(a Space-to-Depth layer followed by a non-strided Convolution,SPD-Conv)模块;其次,在Neck层融合特征时增加上下文转换自注意力(Contextual Transformer,CoT)模块提高模型特征表达能力;最后,引入SIoU,加快模型的收敛速度并提高准确率。所提方法在KITTI数据集上实验。结果显示:相较于原YOLOv8算法,所提算法的准确率、召回率、平均准确率分别提高0.7%、2.1%、2.1%,浮点运算数、帧率分别提高3.6 GFLOPS、24.64 frame/s,证明所提方法能够有效综合满足自动驾驶车辆行人及车辆检测任务中的实时性、精度提高以及降低漏检率和误检率需求。展开更多
In this paper,the authors employ the splitting method to address support vector machine within a reproducing kernel Banach space framework,where a lower semi-continuous loss function is utilized.They translate support...In this paper,the authors employ the splitting method to address support vector machine within a reproducing kernel Banach space framework,where a lower semi-continuous loss function is utilized.They translate support vector machine in reproducing kernel Banach space with such a loss function to a finite-dimensional tensor optimization problem and propose a splitting method based on the alternating direction method of mul-tipliers.Leveraging Kurdyka-Lojasiewicz property of the augmented Lagrangian function,the authors demonstrate that the sequence derived from this splitting method is globally convergent to a stationary point if the loss function is lower semi-continuous and subana-lytic.Through several numerical examples,they illustrate the effectiveness of the proposed splitting algorithm.展开更多
为了提高多视图深度估计结果精度,提出一种基于自适应空间特征增强的多视图深度估计算法。设计了由改进后的特征金字塔网络(feature pyramid network,FPN)和自适应空间特征增强(adaptive space feature enhancement,ASFE)组成的多尺度...为了提高多视图深度估计结果精度,提出一种基于自适应空间特征增强的多视图深度估计算法。设计了由改进后的特征金字塔网络(feature pyramid network,FPN)和自适应空间特征增强(adaptive space feature enhancement,ASFE)组成的多尺度特征提取模块,获取到具有全局上下文信息和位置信息的多尺度特征图像。通过残差学习网络对深度图进行优化,防止多次卷积操作出现重建边缘模糊的问题。通过分类的思想构建focal loss函数增强网络模型的判断能力。由实验结果可知,该算法在DTU(technical university of denmark)数据集上和CasMVSNet(Cascade MVSNet)算法相比,在整体精度误差、运行时间、显存资源占用上分别降低了14.08%、72.15%、4.62%。在Tanks and Temples数据集整体评价指标Mean上该模型优于其他算法,证明提出的基于自适应空间特征增强的多视图深度估计算法的有效性。展开更多
基金W.-X.Li's research was supported by NSF of China(11871054,11961160716,12131017)the Natural Science Foundation of Hubei Province(2019CFA007)T.Yang's research was supported by the General Research Fund of Hong Kong CityU(11304419).
文摘We consider a Prandtl model derived from MHD in the Prandtl-Hartmann regime that has a damping term due to the effect of the Hartmann boundary layer.A global-in-time well-posedness is obtained in the Gevrey function space with the optimal index 2.The proof is based on a cancellation mechanism through some auxiliary functions from the study of the Prandtl equation and an observation about the structure of the loss of one order tangential derivatives through twice operations of the Prandtl operator.
文摘针对自动驾驶场景动态目标检测存在检测速度难以满足实时性要求、检测目标小或被遮挡造成的精度不足和误检、漏检率高等问题,提出一种基于改进YOLOv8模型的行人及车辆检测方法。首先,在Backbone骨干网络提取图像特征时使用对图像分辨率低、小目标检测友好的空间到深度卷积(a Space-to-Depth layer followed by a non-strided Convolution,SPD-Conv)模块;其次,在Neck层融合特征时增加上下文转换自注意力(Contextual Transformer,CoT)模块提高模型特征表达能力;最后,引入SIoU,加快模型的收敛速度并提高准确率。所提方法在KITTI数据集上实验。结果显示:相较于原YOLOv8算法,所提算法的准确率、召回率、平均准确率分别提高0.7%、2.1%、2.1%,浮点运算数、帧率分别提高3.6 GFLOPS、24.64 frame/s,证明所提方法能够有效综合满足自动驾驶车辆行人及车辆检测任务中的实时性、精度提高以及降低漏检率和误检率需求。
基金supported by the National Natural Science Foundation of China(Nos.12026602,12071157,12271108)the Natural Science Foundation of Guangdong Provience(No.2024A1515012288)+1 种基金the Science and Technology Commission of Shanghai Municipality(No.23JC1400501)the Ministry of Science and Technology of China(No.G2023132005L).
文摘In this paper,the authors employ the splitting method to address support vector machine within a reproducing kernel Banach space framework,where a lower semi-continuous loss function is utilized.They translate support vector machine in reproducing kernel Banach space with such a loss function to a finite-dimensional tensor optimization problem and propose a splitting method based on the alternating direction method of mul-tipliers.Leveraging Kurdyka-Lojasiewicz property of the augmented Lagrangian function,the authors demonstrate that the sequence derived from this splitting method is globally convergent to a stationary point if the loss function is lower semi-continuous and subana-lytic.Through several numerical examples,they illustrate the effectiveness of the proposed splitting algorithm.
文摘为了提高多视图深度估计结果精度,提出一种基于自适应空间特征增强的多视图深度估计算法。设计了由改进后的特征金字塔网络(feature pyramid network,FPN)和自适应空间特征增强(adaptive space feature enhancement,ASFE)组成的多尺度特征提取模块,获取到具有全局上下文信息和位置信息的多尺度特征图像。通过残差学习网络对深度图进行优化,防止多次卷积操作出现重建边缘模糊的问题。通过分类的思想构建focal loss函数增强网络模型的判断能力。由实验结果可知,该算法在DTU(technical university of denmark)数据集上和CasMVSNet(Cascade MVSNet)算法相比,在整体精度误差、运行时间、显存资源占用上分别降低了14.08%、72.15%、4.62%。在Tanks and Temples数据集整体评价指标Mean上该模型优于其他算法,证明提出的基于自适应空间特征增强的多视图深度估计算法的有效性。
