Face anti-spoofing (FAS) plays a vital role in face recognition systems. Most state-of-the-art FAS methods 1) rely on stacked convolutions and expert-designed network, which is weak describing detailed fine-grained information easily being ineffective when the environment varies (e.g., different illumination), 2) prefer to use long sequence as input extract dynamic features, making them difficult deploy into scenarios need quick response. Here we propose novel frame level method based...

Recently, deep Convolutional Neural Networks (CNNs) can achieve human-level performance in edge detection with the rich and abstract representation capacities. However, high of CNN based is achieved a large pretrained backbone, which memory energy consuming. In addition, it surprising that previous wisdom from traditional detectors, such as Canny, Sobel, LBP are rarely investigated rapid-developing learning era. To address these issues, we propose simple, lightweight yet effective...

Recently, there have been tremendous efforts in developing lightweight Deep Neural Networks (DNNs) with satisfactory accuracy, which can enable the ubiquitous deployment of DNNs edge devices. The core challenge compact and efficient lies how to balance competing goals achieving high accuracy efficiency. In this paper we propose two novel types convolutions, dubbed \emph{Pixel Difference Convolution (PDC) Binary PDC (Bi-PDC)} enjoy following benefits: capturing higher-order local differential...

Moving object detection in satellite videos (SVMOD) is a challenging task due to the extremely dim and small target characteristics. Current learning-based methods extract spatio-temporal information from multi-frame dense representation with labor-intensive manual labels tackle SVMOD, which needs high annotation costs contains tremendous computational redundancy severe imbalance between foreground background regions. In this paper, we propose highly efficient unsupervised framework for...

Existing Cross-Domain Few-Shot Learning (CDFSL) methods require access to source domain data train a model in the pre-training phase. However, due increasing concerns about privacy and desire reduce transmission training costs, it is necessary develop CDFSL solution without accessing data. For this reason, paper explores Source-Free (SF-CDFSL) problem, which addressed through use of existing pretrained models instead with data, avoiding lack we face two key challenges: effectively tackling...

Binary neural networks (BNNs) constrain weights and activations to +1 or -1 with limited storage computational cost, which is hardware-friendly for portable devices. Recently, BNNs have achieved remarkable progress been adopted into various fields. However, the performance of sensitive activation distribution. The existing utilized Sign function predefined learned static thresholds binarize activations. This process limits representation capacity since different samples may adapt unequal...

Efficiency and robustness are increasingly needed for applications on 3D point clouds, with the ubiquitous use of edge devices in scenarios like autonomous driving robotics, which often demand real-time reliable responses. The paper tackles challenge by designing a general framework to construct learning architectures SO(3) equivariance network binarization. However, naive combination equivariant networks binarization either causes sub-optimal computational efficiency or geometric ambiguity....

Research of the clothing recommendation algorithm is important that can be used to provide a more efficient method for consumers select their expected clothing. Considering characteristics product, in this paper, personalized based on fine-grained attributes reported. In method, are established image. And preference model each user combining with and personal parameters built. This an application system client/server framework mobile phone software Android platform.

Face anti-spoofing (FAS) plays a vital role in face recognition systems. Most state-of-the-art FAS methods 1) rely on stacked convolutions and expert-designed network, which is weak describing detailed fine-grained information easily being ineffective when the environment varies (e.g., different illumination), 2) prefer to use long sequence as input extract dynamic features, making them difficult deploy into scenarios need quick response. Here we propose novel frame level method based...

Recently, deep Convolutional Neural Networks (CNNs) can achieve human-level performance in edge detection with the rich and abstract representation capacities. However, high of CNN based is achieved a large pretrained backbone, which memory energy consuming. In addition, it surprising that previous wisdom from traditional detectors, such as Canny, Sobel, LBP are rarely investigated rapid-developing learning era. To address these issues, we propose simple, lightweight yet effective...

Face perception is an essential and significant problem in pattern recognition, concretely including Recognition (FR), Facial Expression (FER), Race Categorization (RC). Though handcrafted features perform well on face images, Deep Convolutional Neural Networks (DCNNs) have brought new vitality to this field recently. Vanilla DCNNs are powerful at learning high-level semantic features, but weak capturing low-level image characteristic changes illumination, intensity,and texture regarded as...

Replacing normal convolutions with group can significantly increase the computational efficiency of modern deep convolutional networks, which has been widely adopted in compact network architecture designs. However, existing undermine original structures by cutting off some connections permanently resulting significant accuracy degradation. In this paper, we propose dynamic convolution (DGC) that adaptively selects part input channels to be connected within each for individual samples on...

As is well-known, defects precisely affect the lives and functions of machines in which they occur, even cause potentially catastrophic casualties. Therefore, quality assessment before mounting an indispensable requirement for factories. Apart from recognition accuracy, current networks suffer excessive computing complexity, making it great difficulty to deploy manufacturing process. To address these issues, this paper introduces binary into area surface defect detection first time, reason...

This article proposes a novel module called middle spectrum grouped convolution (MSGC) for efficient deep convolutional neural networks (DCNNs) with the mechanism of convolution. It explores broad "middle spectrum" area between channel pruning and conventional Compared pruning, MSGC can retain most information from input feature maps due to group mechanism; compared convolution, benefits learnability, core constructing its topology, leading better division. The is unfolded along four...

Despite recent advancements in high-fidelity human reconstruction techniques, the requirements for densely captured images or time-consuming per-instance optimization significantly hinder their applications broader scenarios. To tackle these issues, we present HumanSplat which predicts 3D Gaussian Splatting properties of any from a single input image generalizable manner. In particular, comprises 2D multi-view diffusion model and latent transformer with structure priors that adeptly...

In this paper, we present a novel 3D head avatar creation approach capable of generalizing from few-shot in-the-wild data with high-fidelity and animatable robustness. Given the underconstrained nature problem, incorporating prior knowledge is essential. Therefore, propose framework comprising learning phases. The phase leverages priors derived large-scale multi-view dynamic dataset, applies these for personalization. Our effectively captures by utilizing Gaussian Splatting-based...

PM2.5 is an important indicator of the severity air pollution and its level can be predicted through hazy photographs caused by degradation. Image-based estimation thus extensively employed in various multimedia applications but challenging because ill-posed property. In this paper, we convert it to problem estimating PM2.5-relevant haze transmission propose a learning model called filtering network. Different from most methods that generate map directly image, our takes coarse derived dark...

The paper analyzed the influence of friction factor theoretically on brake system to produce noise, through complex modal analysis method, established finite element model air disc analyze and forecast noise get noises frequency a certain test conditions. Through multiple sets under different coefficient, it is concluded that increase coefficient has promoting effect noise.