Search

Crowd localization can provide the positions of individuals and the total number of people, which has great application value for security monitoring and public management, meanwhile it meets the challenges of lighting, occlusion and perspective effect. In recent times, Transformer has been applied in crowd localization to overcome these challenges. Yet such kind of methods only consider to integrate the multi-scale information once, which results in incomplete multi-scale information fusion. In this paper, we propose a novel Transformer network named Cross-scale Vision Transformer (CsViT) for crowd localization, which simultaneously fuses multi-scale information during both the encoder and decoder stages and meanwhile building the long-range context dependencies on the combined feature maps. To this end, we design the multi-scale encoder to fuse the feature maps of multiple scales at corresponding positions so as to obtain the combined feature maps, and meanwhile design the multi-scale decoder to integrate the tokens at multiple scales when modeling the long-range context dependencies. Furthermore, we propose Multi-scale SSIM (MsSSIM) loss to adaptively compute head regions and optimize the similarity at multiple scales. Specifically, we set the adaptive windows with different scales for each head and compute the loss values within these windows so as to enhance the accuracy of the predicted distance transform map. We perform comprehensive experiments on five public datasets, and the results obtained validate the effectiveness of our method.

Abstract Cross‐modality person re‐identification (Re‐ID) aims to retrieve a query identity from red, green, blue (RGB) images or infrared (IR) images. Many approaches have been proposed to reduce the distribution gap between RGB modality and IR modality. However, they ignore the valuable collaborative relationship between RGB modality and IR modality. Hybrid Mutual Learning (HML) for cross‐modality person Re‐ID is proposed, which builds the collaborative relationship by using mutual learning from the aspects of local features and triplet relation. Specifically, HML contains local‐mean mutual learning and triplet mutual learning where they focus on transferring local representational knowledge and structural geometry knowledge so as to reduce the gap between RGB modality and IR modality. Furthermore, Hierarchical Attention Aggregation is proposed to fuse local feature maps and local feature vectors to enrich the information of the classifier input. Extensive experiments on two commonly used data sets, that is, SYSU‐MM01 and RegDB verify the effectiveness of the proposed method.

Massive and diverse remote sensing data provide opportunities for data-driven tasks in the real world, but also present challenges in terms of data processing and analysis, especially pixel-level image interpretation. However, the existing shallow-learning and deep-learning segmentation methods, bounded by their technical bottlenecks, cannot properly balance accuracy and efficiency, and are thus hardly scalable to the practice scenarios of remote sensing in a successful way. Instead of following the time-consuming deep stacks of local operations as most state-of-the-art segmentation networks, we propose a novel segmentation model with the encoder–decoder structure, dubbed XANet, which leverages the more computationally economical attention mechanism to boost performance. Two novel attention modules in XANet are proposed to strengthen the encoder and decoder, respectively, namely the Attention Recalibration Module (ARM) and Attention Fusion Module (AFM). Unlike current attention modules, which only focus on elevating the feature representation power, and regard the spatial and channel enhancement of a feature map as two independent steps, ARM gathers element-wise semantic descriptors coupling spatial and channel information to directly generate a 3D attention map for feature enhancement, and AFM innovatively utilizes the cross-attention mechanism for the sufficient spatial and channel fusion of multi-scale features. Extensive experiments were conducted on ISPRS and GID datasets to comprehensively analyze XANet and explore the effects of ARM and AFM. Furthermore, the results demonstrate that XANet surpasses other state-of-the-art segmentation methods in both model performance and efficiency, as ARM yields a superior improvement versus existing attention modules with a competitive computational overhead, and AFM achieves the complementary advantages of multi-level features under the sufficient consideration of efficiency.

Characters, as a kind of symbols carrying rich semantic information, are composed of strokes arranged in a certain structure and are of great significance in our daily life. In this paper, we are concerned with the problem of scene character recognition, and study the problem from the perspective of feature representation. We propose a novel pooling method termed deep contextual stroke pooling (DCSP) for scene character recognition. The proposed DCSP discovers the most prominent stroke information by using stroke detectors and captures the spatial context of discriminative strokes by learning contextual factor. Specifically, we first utilize the convolutional summing map in one convolutional layer to select discriminative strokes and use the convolutional activation features of discriminative strokes to train stroke detectors. Then, we propose the contextual factor to represent the co-occurrence probability of the stroke and its location. Finally, in the response regions, we incorporate the contextual factor into the detector scores and obtain the deep contextual confidence vectors of scene characters. Extensive experiments are conducted on three databases, i.e., ICDAR2003, Chars74k, and SVHN, and the experimental results demonstrate that our method achieves higher accuracies than the state-of-the-art methods.

Driven by the rapid growth of communication technologies and the wide applications of intelligent mobile terminals, the scene character recognition has become a significant yet very challenging task in people's lives. In this paper, we design a novel feature representation scheme termed consecutive convolutional activations (CCA) for character recognition in natural scenes. The proposed CCA could integrate both the low-level and the high-level patterns into the global decision by learning character representations from several successive convolutional layers. Concretely, one shallow convolutional layer is first selected for extracting the convolutional activation features, and then, the next consecutive deep convolutional layers are utilized to learn weight matrices for these convolutional activation features. Finally, the Fisher vectors are employed to encode the CCA features so as to obtain the image-level representations. Extensive experiments are conducted on two English scene character databases (ICDAR2003 and Chars74K) and one Chinese scene character database (“Pan+ChiPhoto”), and the experimental data indicate that the proposed method achieves a superior performance than the previous algorithms.

Coastal aquaculture areas are some of the main areas to obtain marine fishery resources and are vulnerable to storm-tide disasters. Obtaining the information of coastal aquaculture areas quickly and accurately is important for the scientific management and planning of aquaculture resources. Recently, deep neural networks have been widely used in remote sensing to deal with many problems, such as scene classification and object detection, and there are many data sources with different spatial resolutions and different uses with the development of remote sensing technology. Thus, using deep learning networks to extract coastal aquaculture areas often encounters the following problems: (1) the difficulty in labeling; (2) the poor robustness of the model; (3) the spatial resolution of the image to be processed is inconsistent with that of the existing samples. In order to fix these problems, this paper proposes a novel semi-/weakly-supervised method, the semi-/weakly-supervised semantic segmentation network (Semi-SSN), and adopts 3 data sources: GaoFen-2 image, GaoFen-1(PMS)image, and GanFen-1(WFV)image with a 0.8 m, 2 m, and 16 m spatial resolution, respectively, and through experiments, we analyze the extraction effect of the model comprehensively. After comparing with other the-state-of-art methods and verifying on an open remote sensing dataset, we take the Fujian coastal area (mainly Sanduo) as the experimental area and employ our method to detect the effect of storm-tide disasters on coastal aquaculture areas, monitor the production, and make the distribution map of coastal aquaculture areas.

In recent times, deep neural networks have drawn much attention in ground-based cloud recognition. Yet such kind of approaches simply center upon learning global features from visual information, which causes incomplete representations for ground-based clouds. In this paper, we propose a novel method named multi-evidence and multi-modal fusion network (MMFN) for ground-based cloud recognition, which could learn extended cloud information by fusing heterogeneous features in a unified framework. Namely, MMFN exploits multiple pieces of evidence, i.e., global and local visual features, from ground-based cloud images using the main network and the attentive network. In the attentive network, local visual features are extracted from attentive maps which are obtained by refining salient patterns from convolutional activation maps. Meanwhile, the multi-modal network in MMFN learns multi-modal features for ground-based cloud. To fully fuse the multi-modal and multi-evidence visual features, we design two fusion layers in MMFN to incorporate multi-modal features with global and local visual features, respectively. Furthermore, we release the first multi-modal ground-based cloud dataset named MGCD which not only contains the ground-based cloud images but also contains the multi-modal information corresponding to each cloud image. The MMFN is evaluated on MGCD and achieves a classification accuracy of 88.63% comparative to the state-of-the-art methods, which validates its effectiveness for ground-based cloud recognition.

View variation is a major challenge in face recognition. In this study, the authors propose a novel cross‐view face recognition method by seeking potential intermediate domains between the source and target views to model the connection of varying‐views faces. Specifically, each intermediate domain is associated with a dictionary subspace. Learning proceeds in two phases. First, the authors discriminatively train a sub‐dictionary for each subclass of data, which then compose a structured dictionary of powerful reconstructive and discriminative capability on the source data. Secondly, the authors gradually adapt the source domain dictionary to the target domain by incrementally reducing the reconstruction error on the target data, which forms a smooth transition path connecting the source and target domains. Instead of updating the structured dictionary integrally, the authors develop a refined sub‐dictionary‐based updating algorithm, which makes the intermediate dictionaries fit on the target data better and faster. Finally, the authors apply invariant sparse codes across the source, intermediate and target domains to render domain‐shared representations, where the sample differences caused by view changes are reduced. Experiments on the CMU‐PIE and Multi‐PIE dataset demonstrate the effectiveness of the proposed method.

Text information contained in scene images is very helpful for high‐level image understanding. In this study, the authors propose to learn co‐occurrence of local strokes for scene text recognition by using a spatiality embedded dictionary (SED). Unlike spatial pyramid partitioning images into grids to incorporate spatial information, the authors SED associates every codeword with a particular response region and introduces more precise spatial information for robust character recognition. After localised soft coding and max pooling of the first layer, a sparse dictionary is learned to model co‐occurrence of several local strokes, which further improves classification performance. Experimental results on two scene character recognition datasets ICDAR2003 and CHARS74 K demonstrate that their character recognition method outperforms state‐of‐the‐art methods. Besides, competitive word recognition results are also reported for four benchmark word recognition datasets ICDAR2003, ICDAR2011, ICDAR2013 and street view text when combining their character recognition method with a conditional random field language model.

Pooling strategies, such as max pooling and sum pooling, have been widely used to obtain the global representations for action videos. However, these pooling strategies have several disadvantages. First, they are easily affected by unwanted background local features, the absence of discriminative local features and the times of actions periodically performed by actors. Second, most pooling strategies only use local features to build the global representation that captures little mid‐level features for action representation. In this study, the authors propose a novel weighted pooling strategy based on actionlets representation for action recognition. The actionlets are defined as the movements of large bodies such as legs, arms and head, which capture rich mid‐level features for action representation. Besides, the authors’ method also incorporates the distribution information of actionlets into pooling procedure. Specifically, a pooling weight, which determines the importance of actionlet on the final video representation, is assigned to each actionlet. To learn the weight, they propose a novel discriminative learning algorithm to capture the discriminative information for pooling operation. They evaluate their weighted pooling on three datasets: KTH actions dataset, UCF sports dataset and Youtube actions dataset. Experimental results show the effectiveness of the proposed method.

The accurate ground-based cloud classification is a challenging task and still under development. The most current methods are limited to only taking the cloud visual features into consideration, which is not robust to the environmental factors. In this paper, we present the novel joint fusion convolutional neural network (JFCNN) to integrate the multimodal information for ground-based cloud classification. To learn the heterogeneous features (visual features and multimodal features) from the ground-based cloud data, we designed the proposed JFCNN as a two-stream structure which contains the vision subnetwork and multimodal subnetwork. We also proposed a novel layer named joint fusion layer to jointly learn two kinds of cloud features under one framework. After training the proposed JFCNN, we extracted the visual and multimodal features from the two subnetworks and integrated them using a weighted strategy. The proposed JFCNN was validated on the multimodal ground-based cloud (MGC) dataset and achieved remarkable performance, demonstrating its effectiveness for ground-based cloud classification task.

Since cloud images captured from different views possess extreme variations, multi-view ground-based cloud recognition is a very challenging task. In this paper, a study of view shift is presented in this field. We focus both on designing proper feature representation and learning distance metrics from sample pairs. Correspondingly, we propose transfer deep local binary patterns (TDLBP) and weighted metric learning (WML). On one hand, to deal with view shift, like variations of illuminations, locations, resolutions and occlusions, we first utilize cloud images to train a convolutional neural network (CNN), and then extract local features from the part summing maps (PSMs) based on feature maps. Finally, we maximize the occurrences of regions for the final feature representation. On the other hand, the number of cloud images in each category varies greatly, leading to the unbalanced similar pairs. Hence, we propose a weighted strategy for metric learning. We validate the proposed method on three cloud datasets (the MOC_e, IAP_e, and CAMS_e) that are collected by different meteorological organizations in China, and the experimental results show the effectiveness of the proposed method.

Cross-domain ground-based cloud classification is a challenging issue as the appearance of cloud images from different cloud databases possesses extreme variations. Two fundamental problems which are essential for cross-domain ground-based cloud classification are feature representation and similarity measurement. In this paper, we propose an effective feature representation called transfer of local features (TLF), and measurement method called discriminative metric learning (DML). The TLF is a generalized representation framework that can integrate various kinds of local features, e.g., local binary patterns (LBP), local ternary patterns (LTP) and completed LBP (CLBP). In order to handle domain shift, such as variations of illumination, image resolution, capturing location, occlusion and so on, the TLF mines the maximum response in regions to make a stable representation for domain variations. We also propose to learn a discriminant metric, simultaneously. We make use of sample pairs and the relationship among cloud classes to learn the distance metric. Furthermore, in order to improve the practicability of the proposed method, we replace the original cloud images with the convolutional activation maps which are then applied to TLF and DML. The proposed method has been validated on three cloud databases which are collected in China alone, provided by Chinese Academy of Meteorological Sciences (CAMS), Meteorological Observation Centre (MOC), and Institute of Atmospheric Physics (IAP). The classification accuracies outperform the state-of-the-art methods.