Fast adaptation of deep neural networks (DNN) is an important research topic in learning. In this paper, we have proposed a general scheme for DNN based on discriminant condition codes, which are directly fed to various layers pre-trained through new set connection weights. Moreover, present several training methods learn weights from data as well the corresponding code each test condition. work, fast applied supervised speaker speech recognition either frame-level cross-entropy or...

A locally linear K Nearest Neighbor (LLK) method is presented in this paper with applications to robust visual recognition. Specifically, the concept of an ideal representation first presented, which improves upon traditional sparse many ways. The objective function based on a host criteria for sparsity, locality, and reconstruction then optimized derive novel representation, approximation representation. further processed by two classifiers, namely, LLK-based classifier nearest mean-based...

This thesis is based on the application of Internet Things (IoT) and WebGIS in precision agriculture. Through analyzing current development agriculture China considering its advantages shortcomings, we choose an ecology farm as example to conduct a new management system (PAMS) above two techniques. We designed four architectures PAMS: spatial information infrastructure platform, IoT platform mobile client. Users can monitor manage production by PAMS. What's more, module integration method...

Most image segmentation methods based on clustering algorithms use single-objective function to implement segmentation. To avoid the defect, this paper proposes a new method multi-objective... | Find, read and cite all research you need Tech Science Press

An innovative inheritable Fisher vector feature (IFVF) method is presented in this paper for kinship verification. Specifically, first derived each image by aggregating the densely sampled SIFT features opponent color space. Second, a new transformation, which maximizes similarity between images while minimizes that non-kinship pair simultaneously, learned based on vectors. As result, IFVF applying transformation image. Finally, novel fractional power cosine measure, shows its theoretical...

This paper presents a novel locally linear KNN model with the goal of not only developing efficient representation and classification methods, but also establishing relation between them so as to approximate some rules, e.g. Bayes decision rule. Towards that end, first, proposed represents test sample combination all training samples derives new by learning coefficients considering reconstruction, locality sparsity constraints. The theoretical analysis shows has grouping effect nearest...

Anthropology studies show that genetic features are inherited by children from their parents resulting in visual resemblance between them. This paper presents a novel SIFT flow based Fisher vector feature (SF-GFVF) which enhances the facial for kinship verification. The proposed SF-GFVF is derived applying similarity enhancement method on and learning an inheritable transformation so as to enhance encode of parent child image relations. In particular, first presented algorithm densely...

From a statistical perspective, the conventional minimum mean squared error (MMSE) criterion can be considered as maximum likelihood (ML) solution under an assumed homoscedastic Gaussian model. However, in this paper, analysis reveals super-Gaussian and heteroscedastic properties of prediction errors nonlinear regression deep neural network (DNN)-based speech enhancement when estimating clean log-power spectral (LPS) components at DNN outputs with noisy LPS features input vectors....

This paper presents a novel general k nearest neighbour classifier (GKNNc) and mean (GNMc) for visual classification. Instead of treating the data equally, both GKNNc GNMc assign weight coefficient to each data. To achieve good performance, conditions properties coefficients are further analysed. Then sparse representation based method is proposed derive GNMc. Experimental results on several representative sets, such as Caltech 101 dataset MIT-67 indoor scenes demonstrate feasibility methods.

The third Pixel-level Video Understanding in the Wild (PVUW CVPR 2024) challenge aims to advance state of art video understanding through benchmarking Panoptic Segmentation (VPS) and Semantic (VSS) on challenging videos scenes introduced large-scale (VIPSeg) test set Scene Parsing (VSPW) set, respectively. This paper details our research work that achieved 1st place winner PVUW'24 VPS challenge, establishing results all metrics, including Quality (VPQ) Tracking (STQ). With minor fine-tuning...

This paper proposes a novel deep learning architecture for semantic segmentation. The proposed Global and Selective Attention Network (GSANet) features Atrous Spatial Pyramid Pooling (ASPP) with sparsemax global attention selective that deploys condensation diffusion mechanism to aggregate the multi-scale contextual information from extracted features. A decoder is also process GSA-ASPP outputs optimizing softmax volume. We are first benchmark performance of segmentation networks...

Image semantic segmentation is ubiquitously used in consumer electronics, such as AI Camera, which require high accuracy at the boundaries between classes. To improve boundary accuracy, we propose low complexity deep-guidance decoder (DGD) networks, trained with a novel learning (SBL) strategy. Our ablation studies on Cityscapes and ADE20K most-frequent 31 classes, when using different encoders feature extractors, confirm effectiveness of our approach. We show that proposed DGD SBL...

This paper presents two novel discriminative dictionary learning models for sparse representation, namely the Fisher model (FDSM) and marginal (MFDSM). To learn FDSM MFDSM efficiently homogeneously, a general regularized is further derived so that both of them can be learned without much modification. Experimental results on four popular databases, extended Yale face database B, AR database, 15 scenes dataset MIT-67 indoor show proposed method improve upon other methods.

An abstract is not available for this content so a preview has been provided. As you have access to content, full PDF via the ‘Save PDF’ action button.

This paper presents an enhanced sparse coding method by exploiting both the generative and discriminative information in representation model. Specifically, proposed (GDSR) integrates two new criteria, namely a criterion criterion, into conventional criterion. The reveals class conditional probability of each dictionary item using distribution coefficients which are derived representing as linear combination training samples. To further enhance ability method, is also applied localized...

This paper presents a novel multiple anthropological Fisher kernel (MAFK) framework for kinship verification. The proposed MAFK framework, which goes beyond the Mahalanobis distance metric learning, integrates anthropology inspired features and derives semantically meaningful similarities between images. major novelty of this comes from following three aspects. First, new (AIF) are derived by extracting AIF-SIFT, AIF-WLD AIF-DAISY on images that enhanced an similarity enhancement method...

This paper proposes a novel deep learning architecture for semantic segmentation. The proposed Global and Selective Attention Network (GSANet) features Atrous Spatial Pyramid Pooling (ASPP) with sparsemax global attention selective that deploys condensation diffusion mechanism to aggregate the multi-scale contextual information from extracted features. A decoder is also process GSA-ASPP outputs optimizing softmax volume. We are first benchmark performance of segmentation networks...

Image semantic segmentation is ubiquitously used in scene understanding applications, such as AI Camera, which require high accuracy and efficiency. Deep learning has significantly advanced the state-of-the-art segmentation. However, many of recent works only consider class ignore accuracies at boundaries between classes. To improve boundary accuracy, we propose low complexity Guided Decoder (DGD) networks, trained with a novel Semantic Boundary-Aware Learning (SBAL) strategy. Our ablation...

In total, there are over 1,200 hanging and hand scroll landscape paintings from five dynasties period to North Song dynasty which have been recorded in various painting histories, treatises, historical records. this study, we selected more than 120 as experimental samples another 240 South supporting materials. The represent 30 categories of motifs, such Snowy Scene, Wintry Forests, Dwellings Autumnal Mountains, the Mountain Stream, etc. image capture technology is used on with two easily...

Deep neural networks (DNNs) can achieve high accuracy when there is abundant training data that has the same distribution as test data. In practical applications, deficiency often a concern. For classification tasks, lack of enough labeled images in set results overfitting. Another issue mismatch between and domains, which poor model performance. This calls for need to have robust efficient deep learning models. this work, we propose approach called Multi-Expert Adversarial Regularization...