Authors
Xuebin Qin, Zichen Zhang, Chenyang Huang, Masood Dehghan, Osmar R. Zaiane, Martin Jagersand
University of Alberta
Portals
Summary
In this paper, we proposed a novel deep network: U2-Net, for salient object detection. The main architecture of our U2-Net is a two-level nested U-structure. The nested Ustructure with our newly designed RSU blocks enables the network to capture richer local and global information from both shallow and deep layers regardless of the resolutions.
Abstract
In this paper, we design a simple yet powerful deep network architecture, U2-Net, for salient object detection (SOD). The architecture of our U2-Net is a two-level nested U-structure. The design has the following advantages: (1) it is able to capture more contextual information from different scales thanks to the mixture of receptive fields of different sizes in our proposed ReSidual U-blocks (RSU), (2) it increases the depth of the whole architecture without significantly increasing the computational cost because of the pooling operations used in these RSU blocks. This architecture enables us to train a deep network from scratch without using backbones from image classification tasks. We instantiate two models of the proposed architecture, U2-Net (176.3 MB, 30 FPS on GTX 1080Ti GPU) and U2-Net† (4.7 MB, 40 FPS), to facilitate the usage in different environments. Both models achieve competitive performance on six SOD datasets.
Related Works
Multi-level deep feature integration; Multi-scale feature extraction