Search

Class activation maps (CAMs) have been widely used on weakly-supervised object localization, which generate attention maps for specific categories in an image. Since CAMs can be obtained using category annotation, which is included in the annotation information of fully-supervised object detection. Therefore, how to adopt attention information in CAMs to improve the performance of fully-supervised object detection is an interesting problem. In this paper, we propose CAM R-CNN to deal with object detection, in which the category-aware attention maps provided by CAMs are integrated into the process of object detection. CAM R-CNN follows the common pipeline of the recent query-based object detectors in an end-to-end fashion, while two key CAM modules are embedded into the process. Specifically, E-CAM module provides embedding-level attention via fusing proposal features and attention information in CAMs with a transformer encoder, and S-CAM module supplies spatial-level attention by multiplying feature maps with the top-activated attention map provided by CAMs. In our experiments, CAM R-CNN demonstrates its superiority compared to several strong baselines on the challenging COCO dataset. Furthermore, we show that S-CAM module can be applied to two-stage detectors such as Faster R-CNN and Cascade R-CNN with consistent gains. [ABSTRACT FROM AUTHOR]

We attempt to reduce the computational costs in vision transformers (ViTs), which increase quadratically in the token number. We present a novel training paradigm that trains only one ViT model at a time, but is capable of providing improved image recognition performance with various computational costs. Here, the trained ViT model, termed super vision transformer (SuperViT), is empowered with the versatile ability to solve incoming patches of multiple sizes as well as preserve informative tokens with multiple keeping rates (the ratio of keeping tokens) to achieve good hardware efficiency for inference, given that the available hardware resources often change from time to time. Experimental results on ImageNet demonstrate that our SuperViT can considerably reduce the computational costs of ViT models with even performance increase. For example, we reduce 2 × FLOPs of DeiT-S while increasing the Top-1 accuracy by 0.2% and 0.7% for 1.5 × reduction. Also, our SuperViT significantly outperforms existing studies on efficient vision transformers. For example, when consuming the same amount of FLOPs, our SuperViT surpasses the recent state-of-the-art EViT by 1.1% when using DeiT-S as their backbones. The project of this work is made publicly available at https://github.com/lmbxmu/SuperViT. [ABSTRACT FROM AUTHOR]

Graph convolutional networks (GCNs) provide a promising way to extract the useful information from graph-structured data. Most of the existing GCNs methods usually focus on local neighborhood information based on specific convolution operations, and ignore the global structure of the input data. To extract the latent representation for the graph-structured data more effectively, we introduce a deepwalk strategy into GCNs to efficiently explore the global graph information. This strategy can complement the local neighborhood information of a graph, resulting in the more robust representation for the graph data. The fusion of the local neighboring and global structured information of a graph can further facilitate deep feature learning at the output layer of GCNs for node classification. Experimental results show that the proposed model has achieved state-of-the-art results on three benchmark datasets including Cora, Citeseer, and Pubmed citation networks. [ABSTRACT FROM AUTHOR]

Consulting through message-updating in social network is regarded as a popular way of information seeking. However, most questions cannot receive answers or suggestions timely, and some questions even fail to get replies. Thus, identifying microblogs that contain questions (we call them 'question microblog') and recommending answers automatically are meaningful. We divide this problem into two submodules: question identification and answer recommendation. To the best of our knowledge, few attempts have been made to identify questions in microblogs due to standard features such as 5W1H (How, What, Where, When, Who, Why) are likely to be absent. The following challenging problem is how to provide users a relevant, credible, diversified and personalized answer after a microblog is recognized as a question. In this paper, we investigate the feasibility of integrating standard features and contextual features extracted from auxiliary resources and recommend a reasonable answer using collaborative filtering. Empirical results on Sina Microblog dataset demonstrate the efficacy and effectiveness of our method. [ABSTRACT FROM AUTHOR]

Human vision system can receive the RGB and depth information at the same time and make an accurate judgment on human behaviors. However, in an ordinary camera, there is a loss in information when a 3D image is projected to a 2D plane. The depth and RGB information collected simultaneously by Kinect can provide more discriminant information for human behaviors than traditional cameras. Therefore, RGB-D camera is thought to be the key of solving human behavior recognition for a long time. In this paper, we develop 3D motion scale invariant feature transform for the description of the depth and motion information. It serves as a more effective descriptor for the RGB and depth videos. Hidden Markov Model is utilized for improving the accuracy of human behavior recognition. Experiments show that our framework provides richer information for discriminative point of behavior analysis and obtains better recognition performance. [ABSTRACT FROM AUTHOR]

The application of two-dimensional (2D) vector map faces the security issues of copyright protection, which limit the usage of vector data in many scenarios. Reversible watermarking is a more feasible solution, which aims to restore the original data after watermark extraction. In this paper, high-capacity reversible watermarking scheme of 2D-vector data is proposed based on the idea of iterative embedding. It groups vertices as units for each polyline and selects highly correlated vertex units as cover data. Then the reversible embedding is carried out by iteratively modifying the median vertex coordinates of each selected embedding unit. This scheme can strictly recover the original vector data with watermark accurate extraction. Meanwhile, both higher payload capacity and perception invisibility are validated through theoretical analysis and comprehensive experiments. The experimental results show that the proposed reversible watermarking scheme is suitable for 2D-vector data applications where high-precision data are required. [ABSTRACT FROM AUTHOR]

The reversible watermarking technique is suitable for vector maps due to its reversibility after watermark extraction. In this paper, a novel reversible watermarking scheme based on the idea of nonlinear scrambling is proposed. It begins with feature point extraction. To avoid the high-precision vector data being illegally used by unauthorized users, the algorithm nonlinearly scrambles the relative position of feature points. Then based on the proposed reversible embedding, both scrambled feature points and nonfeature points are taken as cover data, the coordinates of which are modified to embed both watermark data and feature point identification data. Finally, combined with the scrambling secret key, the original vector data can be exactly recovered with watermark extraction. Comprehensive experimental results validate that the scheme could effectively prevent the high-precision vector data from being illegally used with maintaining the basic shape of each polyline, simultaneously. [ABSTRACT FROM AUTHOR]