To solve the problem of ship heave motion in harsh marine environments, which affects the positioning accuracy and safety of its robotic arm, this paper adopts a PID parameter optimization method based on Tent cold li...
详细信息
Deep learning methods have achieved remarkable results in the direction of image quality assessment tasks. However, most of the related studies focus only on image unimodal, ignoring the potential advantages that come...
详细信息
Texture defect detection has wide applications in both textiles and industry, where traditional defect detection methods focus on costly manual labeling and defect samples are complex to collect in real-world industri...
详细信息
Orbital angular momentum(OAM)modes provide an additional orthogonal physical dimension,offering transformative potential for enhancing optical communication *** significant progress in mode multiplexing,the developmen...
详细信息
Orbital angular momentum(OAM)modes provide an additional orthogonal physical dimension,offering transformative potential for enhancing optical communication *** significant progress in mode multiplexing,the development of robust communication networks faces persistent challenges,particularly in effectively routing and controlling these multiplexed channels among network *** tackle these dilemmas,we propose a rotatable diffractive neural network(R-DNN)strategy and demonstrate its capability for port-controllable OAM mode *** leveraging the correlation between the orthogonal evolution of OAM modes in free space and phase modulations during propagation,the R-DNN precisely shapes the spatial evolution of mode fields through multiple rotatable phase layers,enabling efficient routing to specific output *** approach exploits the interaction of secondary wavelets with the relative states of the rotatable layers,allowing on-demand control of mode evolution paths and enhancing routing *** a proof of concept,we developed a tri-functional router that successfully directs three OAM modes to individually controllable output *** router achieves an average intermode crosstalk of less than−16.4 dB across three functional states,one-dimensional,two-dimensional,and cross-connected switching,while supporting the routing of 5.85 Tbit/s quadrature phase-shift keying *** results highlight the R-DNN’s effectiveness in achieving precise and controllable OAM mode manipulation,paving the way for advanced applications in mode-multiplexed communication networks and beyond.
Salient object detection(SOD)in RGB and depth images has attracted increasing research *** RGB-D SOD models usually adopt fusion strategies to learn a shared representation from RGB and depth modalities,while few meth...
详细信息
Salient object detection(SOD)in RGB and depth images has attracted increasing research *** RGB-D SOD models usually adopt fusion strategies to learn a shared representation from RGB and depth modalities,while few methods explicitly consider how to preserve modality-specific *** this study,we propose a novel framework,the specificity-preserving network(SPNet),which improves SOD performance by exploring both the shared information and modality-specific ***,we use two modality-specific networks and a shared learning network to generate individual and shared saliency prediction *** effectively fuse cross-modal features in the shared learning network,we propose a cross-enhanced integration module(CIM)and propagate the fused feature to the next layer to integrate cross-level ***,to capture rich complementary multi-modal information to boost SOD performance,we use a multi-modal feature aggregation(MFA)module to integrate the modalityspecific features from each individual decoder into the shared *** using skip connections between encoder and decoder layers,hierarchical features can be fully *** experiments demonstrate that our SPNet outperforms cutting-edge approaches on six popular RGB-D SOD and three camouflaged object detection *** project is publicly available at https://***/taozh2017/SPNet.
Remote photoplethysmography (rPPG) is a non-contact technology that can estimate heart rate using facial video and holds significant potential for health monitoring. Despite the latest deep learning-based rPPG approac...
详细信息
ISBN:
(纸本)9798400708343
Remote photoplethysmography (rPPG) is a non-contact technology that can estimate heart rate using facial video and holds significant potential for health monitoring. Despite the latest deep learning-based rPPG approaches can predict high-quality rPPG signal under similar scenarios, these methods often suffer from degraded performance when encountering variations in subjects, environments, or illumination conditions in target domains. To address this challenge, we propose an uncertainty-guided self-training approach that leverages model uncertainty and periodic priors to enhance generalization across different domains without requiring labels in the target domain. We iteratively update the model using pseudo-labels generated from its own predictions on unlabelled data in the target domain, with varying confidence levels informed by the model’s uncertainty estimation. To achieve this, we modify a standard Convolutional Neural Network (CNN) into a Bayesian Neural Network (BNN) for uncertainty estimation, which guides the assignment of pseudo-labels with varying confidence levels. By employing the adversarially learned periodic priors of rPPG signals shared across domains as regularization terms, we further stabilize the model adaptation process. We evaluate the proposed method on two public datasets (PURE and UBFC-rPPG) across five cross-domain tasks. Experimental results demonstrate improved performance over the baselines, with gains ranging from 60.5% to 97.2%, outperforming existing methods in generalization performance for rPPG-based heart rate measurement.
Recently, federated graph learning has attracted significant attention, as subgraphs of a global graph may often distribute across different institutions and are subject to privacy restrictions. However, inevitable da...
详细信息
Currently,data-driven models of solar activity forecast are investigated extensively by using machine *** model training,it is highly demanded to establish a large database which may contain observations coming from d...
详细信息
Currently,data-driven models of solar activity forecast are investigated extensively by using machine *** model training,it is highly demanded to establish a large database which may contain observations coming from different instruments with different spatio-temporal *** this paper,we employ deep learning models for super-resolution(SR)of magnetogram of Michelson Doppler Imager(MDI)in order to achieve the same spatial resolution of Helioseismic and Magnetic Imager(HMI).First,a generative adversarial network(GAN)is designed to transfer characteristics of MDI onto downscaled HMI,getting low-resolution HMI magnetogram in the same domain as ***,with the paired low-resolution and high-resolution HMI magnetograms,another GAN is trained in a supervised learning way,which consists of two streams,one is for generating high-fidelity image content,the other is explicitly optimized for generating elaborate image ***,these two streams work together to guarantee both high-fidelity and photorealistic super-resolved *** results demonstrate that the proposed method can generate super-resolved magnetograms with perceptual-pleasant visual ***,the best PSNR,LPIPS,RMSE,comparable SSIM and CC are obtained by the proposed *** source code and data set can be accessed via https://***/filterbank/SPSR.
The objective of physics-based differentiable rendering (PBDR) is to propagate gradients between scene parameters and the intensities of image pixels in a manner that is physically correct. The gradients obtained can ...
详细信息
ISBN:
(数字)9781510688780
ISBN:
(纸本)9781510688773
The objective of physics-based differentiable rendering (PBDR) is to propagate gradients between scene parameters and the intensities of image pixels in a manner that is physically correct. The gradients obtained can be applied in optimization algorithms for the reconstruction of 3D geometry or materials, or they can be further propagated into neural network to learn neural representations of the scene. However, applying automatic differentiation techniques directly to the primary rendering process will result in biased gradients, as the rendering integral contains moving high-dimensional discontinuities. Based on how these discontinuities are managed-either implicitly or explicitly-existing PBDR methods can be categorized into two groups: reparameterization methods and boundary sampling methods. Boundary sampling methods need to construct paths that have one segment tangent to the geometry being differentiated in order to estimate a boundary integral to address the discontinuities explicitly. Such paths are usually constructed by sampling the tangent segment first and then extending it to complete the paths for subsequent processing. Fortunately, the number of dimensions in the space composed of such tangent segments is only three. In scenes comprised solely of triangle meshes, the first dimension is used to parameterize all the edges on the mesh, which determines a point on the tangent segment. The remaining two dimensions are used to parameterize the direction of the tangent segment. However, state-of-the-art boundary sampling methods parameterize the first dimension uniformly, which is inefficient because only a small portion of the edges contributes to the boundary integral, resulting in wasted parameter space. In this paper, we parameterize the first dimension by considering both edge length and contributions, thereby allocating more parameter space to important edges. Experiments demonstrate that our methods achieve lower variance gradients in the forward dif
Fault injection attacks represent a type of active, physical attack against cryptographic circuits. Various countermeasures have been proposed to thwart such attacks, however, the design and implementation of which ar...
详细信息
暂无评论