Shape descriptor is a concise yet informative representation that provides 3D object with an identification as member of some category. We have developed deep shape to address challenging issues from ever-growing datasets in areas diverse engineering, medicine, and biology. Specifically, this paper, we novel techniques extract but geometrically new methods defining Eigen-shape Fisher-shape guide the training neural network. Our tends maximize inter-class margin while minimize intra-class...

Pedestrian detection relying on deep convolution neural networks has made significant progress. Though promising results have been achieved standard pedestrians, the performance heavily occluded pedestrians remains far from satisfactory. The main culprits are intra-class occlusions involving other and inter-class caused by objects, such as cars bicycles. These in a multitude of occlusion patterns. We propose an approach for pedestrian with following contributions. First, we introduce novel...

The attention mechanism of the Transformer has advantage extracting feature correlation in long-sequence data and visualizing model. As time-series data, spatial temporal dependencies EEG signals between time points different channels contain important information for accurate classification. So far, Transformer-based approaches have not been widely explored motor-imagery classification visualization, especially lacking general models based on cross-individual validation. Taking model...

Pedestrian detection is an important but challenging problem in computer vision, especially human-centric tasks. Over the past decade, significant improvement has been witnessed with help of handcrafted features and deep features. Here we present a comprehensive survey on recent advances pedestrian detection. First, provide detailed review single-spectral that includes based methods approaches. For methods, extensive approaches find large freedom degrees shape space have better performance....

Complex geometric structural variations of 3D model usually pose great challenges in shape matching and retrieval. In this paper, we propose a high-level feature learning scheme to extract features that are insensitive deformations via novel discriminative deep auto-encoder. First, multiscale distribution is developed for use as input the Then, by imposing Fisher discrimination criterion on neurons hidden layer, auto-encoder learning. Finally, layers from multiple auto-encoders concatenated...

Complex geometric variations of 3D models usually pose great challenges in shape matching and retrieval. In this paper, we propose a novel feature learning method to extract high-level features that are insensitive deformations shapes. Our uses discriminative deep auto-encoder learn deformation-invariant features. First, multiscale distribution is computed used as input the auto-encoder. We then impose Fisher discrimination criterion on neurons hidden layer develop Finally, outputs from...

Detection of double JPEG compression plays an important role in digital image forensics. Some successful approaches have been proposed to detect when the primary and secondary compressions different quantization matrices. However, detecting with same matrix is still a challenging problem. In this paper, effective error-based statistical feature extraction scheme presented solve First, given file decompressed form reconstructed image. An error obtained by computing differences between inverse...

Motivated by analysis of genetical genomics data, we introduce a sparse high dimensional multivariate regression model for studying conditional independence relationships among set genes adjusting possible genetic effects. The precision matrix in the specifies covariate-adjusted Gaussian graph, which presents dependence structure gene expression after confounding effects on are taken into account. We present estimation method using constrained ℓ1 minimization, can be easily implemented...

Recently, deep learning based point cloud descriptors have achieved impressive results in the place recognition task. Nonetheless, due to sparsity of clouds, how extract discriminative local features clouds efficiently form a global descriptor is still challenging problem. In this paper, we propose pyramid transformer network (PPT-Net) learn from for efficient retrieval. Specifically, first develop module that adaptively learns spatial relationship different k-NN neighboring points where...

In recent years, how to strike a good trade-off between accuracy, inference speed, and model size has become the core issue for real-time semantic segmentation applications, which plays vital role in real-world scenarios such as autonomous driving systems drones. this study, we devise novel lightweight network using multi-scale context fusion (MSCFNet) scheme, explores an asymmetric encoder-decoder architecture alleviate these problems. More specifically, encoder adopts some developed...

Spectral clustering is an important method widely used for pattern recognition and image segmentation. Classical spectral algorithms consist of two separate stages: 1) solving a relaxed continuous optimization problem to obtain real matrix followed by 2) applying <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${K}$ </tex-math></inline-formula> -means or rotation round the (i.e., result) into binary called...

Point cloud semantic segmentation is a crucial task in 3D scene understanding. Existing methods mainly focus on employing large number of annotated labels for supervised segmentation. Nonetheless, manually labeling such point clouds the time-consuming. In order to reduce labels, we propose semi-supervised network, named SSPC-Net, where train network by inferring unlabeled points from few points. our method, first partition whole into superpoints and build superpoint graphs mine long-range...

Transformer has been widely used in the field of natural language processing (NLP) with its superior ability to handle long-range dependencies comparison convolutional neural network (CNN) and recurrent (RNN). This correlation is also important for recognition time series signals, such as electroencephalogram (EEG). Currently, commonly EEG classification models are CNN, RNN, deep believe (DBN), hybrid CNN. not recognition. In this study, we constructed multiple Transformer-based motor...

We propose a novel one-step transformer-based person search framework, PSTR, that jointly performs detection and re-identification (re-id) in single architecture. PSTR comprises search-specialized (PSS) module contains encoder-decoder for along with discriminative re-id decoder re-id. The utilizes multi-level supervision scheme shared feature learning also part attention block to encode relationship between different parts of person. further introduce simple multi-scale support across...

How to effectively retrieve desired 3D models with simple queries is a long-standing problem in computer vision community. The model-based approach quite straightforward but nontrivial, since people could not always have the query model available by side. Recently, large amounts of wide-screen electronic devices are prevail our daily lives, which makes sketch-based shape retrieval promising candidate due its simpleness and efficiency. main challenge huge modality gap between sketch shape. In...

Sketch-based 3D shape retrieval, which returns a set of relevant shapes based on users' input sketch queries, has been receiving increasing attentions in both graphics community and vision community. In this work, we address the sketch-based retrieval problem with novel Cross-Domain Neural Networks (CDNN) approach, is further extended to Pyramid (PCDNN) by cooperating hierarchical structure. order alleviate discrepancies between features features, neural network pair that forces identical...

Retrieving 3D shapes with sketches is a challenging problem since 2D and are from two heterogeneous domains, which results in large discrepancy between them. In this paper, we propose to learn barycenters of projections for sketch-based shape retrieval. Specifically, first use deep convolutional neural networks (CNNs) extract features shapes. For shapes, then compute the Wasserstein multiple form barycentric representation. Finally, by constructing metric network, discriminative loss...

Heavy haze results in severe image degradation and thus hampers the performance of visual perception, object detection, etc. On assumption that dehazed binocular images are superior to hazy ones for stereo vision tasks such as 3D detection according fact is a function depth, this paper proposes Binocular dehazing Network (BidNet) aiming at both left right within deep learning framework. Existing methods rely on simultaneously estimating disparity, whereas BidNet does not need explicitly...

The explosive growth of 3D models has led to the pressing demand for an efficient searching system. Traditional model-based search is usually not convenient, since people don't always have model available by side. sketch-based shape retrieval a promising candidate due its simpleness and efficiency. main challenge discrepancy across different domains. In paper, we propose novel deep correlated metric learning (DCML) method mitigate between sketch proposed DCML trains two distinct neural...

Existence of haze significantly degrades visual quality and hence negatively affects the performance surveillance, video analysis, human–machine interaction. To remove from a signal, in this paper, we propose generative adversarial network for removal called HRGAN. HRGAN consists generator discriminator network. A unified jointly estimating transmission maps, atmospheric light, haze-free images (called UNTA) is proposed as Instead being optimized by minimizing pixel-wise loss, novel loss...