Images captured in foggy weather conditions often suffer from bad visibility. In this paper, we propose an efficient regularization method to remove hazes a single input image. Our benefits much exploration on the inherent boundary constraint transmission function. This constraint, combined with weighted L_1-norm based contextual regularization, is modeled into optimization problem estimate unknown scene transmission. A quite algorithm variable splitting also presented solve problem. The...

Point cloud analysis is very challenging, as the shape implied in irregular points difficult to capture. In this paper, we propose RS-CNN, namely, Relation-Shape Convolutional Neural Network, which extends regular grid CNN configuration for point analysis. The key RS-CNN learning from relation, i.e., geometric topology constraint among points. Specifically, convolutional weight local set forced learn a high-level relation expression predefined priors, between sampled and others. way, an...

Image clustering is a crucial but challenging task in machine learning and computer vision. Existing methods often ignore the combination between feature clustering. To tackle this problem, we propose Deep Adaptive Clustering (DAC) that recasts problem into binary pairwise-classification framework to judge whether pairs of images belong same clusters. In DAC, similarities are calculated as cosine distance label features which generated by deep convolutional network (ConvNet). By introducing...

Current state-of-the-art detectors typically exploit feature pyramid to detect objects at different scales. Among them, FPN is one of the representative works that build a by multi-scale features summation. However, design defects behind prevent from being fully exploited. In this paper, we begin first analyzing in FPN, and then introduce new architecture named AugFPN address these problems. Specifically, consists three components: Consistent Supervision, Residual Feature Augmentation, Soft...

Accurate road detection and centerline extraction from very high resolution (VHR) remote sensing imagery are of central importance in a wide range applications. Due to the complex backgrounds occlusions trees cars, most methods bring heterogeneous segments; besides for task, current approaches fail extract wonderful network that appears smooth, complete, as well single-pixel width. To address above-mentioned issues, we propose novel deep model, i.e., cascaded end-to-end convolutional neural...

Point cloud processing is very challenging, as the diverse shapes formed by irregular points are often indistinguishable. A thorough grasp of elusive shape requires sufficiently contextual semantic information, yet few works devote to this. Here we propose DensePoint, a general architecture learn densely representation for point processing. Technically, it extends regular grid CNN configuration generalizing convolution operator, which holds permutation invariance points, and achieves...

Hyperspectral unmixing, the process of estimating a common set spectral bases and their corresponding composite percentages at each pixel, is an important task for hyperspectral analysis, visualization understanding. From unsupervised learning perspective, this problem very challenging---both are unknown, making solution space too large. To reduce space, many approaches have been proposed by exploiting various priors. In practice, these priors would easily lead to some unsuitable solution....

The cross-modal feature matching has gained much attention in recent years, which many practical applications, such as the text-to-image retrieval. most difficult problem of is how to eliminate heterogeneity between modalities. existing methods (e.g., CCA and PLS) try learn a common latent subspace, where two modalities minimized so that cross-matching possible. However, these require fully paired samples suffer difficulties when dealing with unpaired data. Besides, utilizing class label...

Semantic segmentation is a fundamental task in remote sensing image processing. The large appearance variations of ground objects make this quite challenging. Recently, deep convolutional neural networks (DCNNs) have shown outstanding performance task. A common strategy these methods (e.g., SegNet) for improvement to combine the feature maps learned at different DCNN layers. However, such combination usually implemented via map summation or concatenation, indicating that features are...

As an indispensable part in Intelligent Traffic System (ITS), the task of traffic forecasting inherently subjects to following three challenging aspects. First, data are physically associated with road networks, and thus should be formatted as graphs rather than regular grid-like tensors. Second, render strong spatial dependence, which implies that nodes usually have complex dynamic relationships between each other. Third, demonstrate temporal is crucial for time series modeling. To address...

Predicting traffic flow on networks is a very challenging task, due to the complicated and dynamic spatial-temporal dependencies between different nodes network. The renders two types of temporal dependencies, including short-term neighboring long-term periodic dependencies. What's more, spatial correlations over are both local non-local. To capture global correlations, we propose Global Spatial-Temporal Network (GSTNet), which consists several layers blocks. Each block contains...

Super-resolution from a single image plays an important role in many computer vision systems. However, it is still challenging task, especially preserving local edge structures. To construct high-resolution images while the sharp edges, effective edge-directed super-resolution method presented this paper. An adaptive self-interpolation algorithm first proposed to estimate gradient field directly input low-resolution image. The obtained then regarded as constraint or edge-preserving...

This brief presents a framework of retargeted least squares regression (ReLSR) for multicategory classification. The core idea is to directly learn the targets from data other than using traditional zero-one matrix as targets. learned target can guarantee large margin constraint requirement correct classification each point. Compared with (LSR) and recently proposed discriminative LSR models, ReLSR much more accurate in measuring error model. Furthermore, single compact model, hence there no...

Cross-modal retrieval emphasizes understanding inter-modality semantic correlations, which is often achieved by designing a similarity function. Generally, one of the most important things considered function how to make cross-modal computable. In this paper, deep and bidirectional representation learning model proposed address issue image-text retrieval. Owing solid progress in computer vision natural language processing, it reliable extract representations from both raw image text data...

This paper presents local spline regression for semi-supervised classification. The core idea in our approach is to introduce splines developed Sobolev space map the data points directly be class labels. composed of polynomials and Green's functions. It smooth, nonlinear, able interpolate scattered with high accuracy. Specifically, each neighborhood, an optimal estimated via regularized least squares regression. With this spline, neighboring mapped a label. Then, loss evaluated further...

Neural Architecture Search (NAS) is an important yet challenging task in network design due to its high computational consumption. To address this issue, we propose the Reinforced Evolutionary (RENAS), which evolutionary method with reinforced mutation for NAS. Our integrates into evolution algorithm neural architecture exploration, a controller introduced learn effects of slight modifications and make actions. The guides model population evolve efficiently. Furthermore, as child models can...