Deep Superpixel Convolutional Network for Image Recognition

被引:8
|
作者
Zeng, Xianfang [1 ]
Wu, Wenxuan [2 ]
Tian, Guangzhong [1 ]
Li, Fuxin [2 ]
Liu, Yong [3 ]
机构
[1] Zhejiang Univ, Inst Cyber Syst & Control, Hangzhou 310027, Peoples R China
[2] Oregon State Univ, CORIS Inst, Corvallis, OR 97331 USA
[3] Zhejiang Univ, State Key Lab Ind Control Technol, Hangzhou 310027, Peoples R China
基金
中国国家自然科学基金;
关键词
Convolution; Three-dimensional displays; Task analysis; Standards; Image recognition; Kernel; Feature extraction; Deep learning; representation learning; image recognition; superpixel;
D O I
10.1109/LSP.2021.3075605
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
Due to the high representational efficiency, superpixel largely reduces the number of image primitives for subsequent processing. However, superpixel is scarcely utilized in recent methods since its irregular shape is intractable for standard convolutional layer. In this paper, we propose an end-to-end trainable superpixel convolutional network, named SPNet, to learn high-level representation on image superpixel primitives. We start by treating irregular superpixel lattices as a 2D point cloud, where the low-level features inside one superpixel are aggregated to one feature vector. We replace the standard convolutional layer with the PointConv layer to handle the irregular and unordered point cloud. Besides, we propose grid based downsampling strategies to output uniform 2D sampling result. The resulting network largely utilizes the efficiency of superpixel and provides a novel view for image recognition task. Experiments on image recognition task show promising results compared with prominent image classification methods. The visualization of class activation mapping shows great accuracy at object localization and boundary segmentation.
引用
收藏
页码:922 / 926
页数:5
相关论文
empty
未找到相关数据