Unsupervised domain adaptation (UDA) assumes that source and target data are freely available usually trained together to reduce the gap. However, considering privacy inefficiency of transmission, it is impractical in real scenarios. Hence, draws our eyes optimize network without accessing labeled data. To explore this direction object detection, for first time, we propose a data-free adaptive detection (SFOD) framework via modeling into problem learning with noisy labels. Generally,...

Periodic surface structures with periods as small about one-tenth of the irradiating femtosecond (fs) laser light wavelength were created on a titanium (Ti) foil by exploiting laser-induced oxidation and third harmonic generation (THG). They achieved using 100-fs pulses repetition rate 1 kHz ranging from 1.4 to 2.2 μm. It was revealed that an extremely thin TixOy layer formed Ti after fs fluence smaller than ablation threshold Ti, leading significant enhancement in THG which may exceed...

Conventional face anti-spoofing methods might be poorly generalized to unseen data distributions. Thus, we improve the generalization of spoof detection from multi-domain feature disentanglement. Specially, a two-branch convolutional network is proposed separate spoof-specific features and domain-specific images explicitly. The are further used for live vs. classification. To minimize correlation among these two features, present cross-adversarial training scheme, which requires each branch...

Most of the face anti-spoofing methods improve generalization capability by adversarial domain adaptation via training source and target data jointly. However, considering privacy, it is impractical in application. Hence, we propose a data-free adaptative framework to optimize network without using labeled modeling into problem learning with noisy labels. To obtain more reliable pseudo labels, dynamic images background capture motion divergences between real attack faces. Nonetheless,...

This paper reviews the machine-vision techniques available for image acquisition and their processing-analysis in agricultural automation up to now according essential base core work, focusing on 2 types of image, i.e., visible infrared image. Results from previous studies have shown two important topics agriculture application, visual navigation behaviors surveillance. Each topic introduces major applications respective subfields. Furthermore, each one expresses technical discussions,...

Although some metadata standards such as LOM and Dublin Core are defined to offer the base guarantee for education information sharing, there still shortcomings in e-leaming domain, main one being their exclusive focus on property based specifications. Semantic Web technologies allow us enhance these specifications using classes of objects with common attributes. Another shortcoming is that they do not include any domain ontologies - which again can be specified building semantic formalisms....

Unsupervised domain adaptation (UDA) assumes that source and target data are freely available usually trained together to reduce the gap. However, considering privacy inefficiency of transmission, it is impractical in real scenarios. Hence, draws our eyes optimize network without accessing labeled data. To explore this direction object detection, for first time, we propose a data-free adaptive detection (SFOD) framework via modeling into problem learning with noisy labels. Generally,...

Deep-learning-based techniques have been widely adopted for autonomous driving software stacks mass production in recent years, focusing primarily on perception modules, with some work extending this method to prediction modules. However, the downstream planning and control modules are still designed hefty handcrafted rules, dominated by optimization-based methods such as quadratic programming or model predictive control. This results a performance bottleneck systems that corner cases simply...

In this study, we sequenced and analyzed the mitochondrial genome of Satanas beetle, Dynastes satanas Moser, 1909, which was intercepted by Chinese Customs during an attempted smuggling operation in 2022. The complete is 16,973 bp length (GenBank accession number: OQ998898) contains 13 protein-coding genes, 22 tRNAs, 2 rRNAs, a control region 2285 bp. gene order trnQ-trnI-trnM consistent with that other species genus Dynastes. All PCGs are initiated ATN start codon. Seven genes terminate TAA...

As accurate identification of weeds from crops is the prerequisite for precise herbicides spraying, this paper proposes a multi-feature fusion method based on neutral network and D-S evidential theory to improve accuracy weed recognition. Firstly, three kinds single features such as color, shape texture are extracted crop leaves after series image processing. Secondly, classified with each kind feature by output sub-network made an independent evidences construct basic belief assignment....

3D Morphable Model (3DMM) is a statistical tool widely employed in reconstructing face shape. Existing methods are aimed at predicting 3DMM shape parameters with single encoder but suffer from unclear distinction of different attributes. To address this problem, Two-Pathway Encoder-Decoder Network (2PEDN) proposed to regress the identity and expression components via global local pathways. Specifically, each 2D image cropped into details as inputs for corresponding 2PEDN trained predict two...

Although some metadata standards such as LOM and Dublin Core are defined to offer the base guarantee for education information sharing, there still shortcomings in e-leaming domain, main one being their exclusive focus on property based specifications. Semantic Web technologies allow us enhance these specifications using classes of objects with common attributes. Another shortcoming is that they do not include any domain ontologies - which again can be specified building semantic formalisms....

This paper develops an algorithm to effectively explore the advantages of both sparse vector recovery methods and generative model-based for solving compressed sensing problem. The proposed mainly consists two steps. In first step, a network-based projected gradient descent (NPGD) is introduced solve non-convex optimization problem, obtaining preliminary original signal. Then with obtained recovery, l <sub xmlns:mml="http://www.w3.org/1998/Math/MathML"...

Most recognizing occluded faces methods focus on synthetic-occluded for training due to the lack of real-occluded data. However, performance may suffer from degradation since and face images are under different distributions. Hence, it draws our eyes transfer model synthetic real-world domain. In this paper, we propose a source data-free domain adaptive recognition framework optimize network in target via redefining as pseudo labels denoising problem. To obtain reliable labels, train...

We propose a novel joint framework for 3D face reconstruction (3DFR) that integrates facial attribute estimation (FAE) as an auxiliary task. One of the essential problems 3DFR is to extract semantic features (e.g., Big Nose, High Cheekbones, and Asian) from in-the-wild 2D images, which inherently involved with FAE. These two tasks, though heterogeneous, are highly relevant each other. To achieve this, we leverage Convolutional Neural Network shared representations both shape decoder...