U$^2$-Net: Going Deeper with Nested U-Structure for Salient Object Detection

18 May 2020  ·  Xuebin Qin, Zichen Zhang, Chenyang Huang, Masood Dehghan, Osmar R. Zaiane, Martin Jagersand ·

In this paper, we design a simple yet powerful deep network architecture, U$^2$-Net, for salient object detection (SOD). The architecture of our U$^2$-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, U$^2$-Net (176.3 MB, 30 FPS on GTX 1080Ti GPU) and U$^2$-Net$^{\dagger}$ (4.7 MB, 40 FPS), to facilitate the usage in different environments. Both models achieve competitive performance on six SOD datasets. The code is available: https://github.com/NathanUA/U-2-Net. read more

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Saliency Detection DUT-OMRON U2-Net+ MAE 0.06 # 4
{max}Fβ 0.813 # 1
Fwβ 0.731 # 1
Sm 0.837 # 1
relaxFbβ 0.676 # 1
Saliency Detection DUT-OMRON U2-Net MAE 0.054 # 3
Salient Object Detection DUTS-TE U2-Net+ (RSU) {max}Fβ 0.852 # 2
MAE 0.054 # 3
Fwβ 0.763 # 2
Sm 0.847 # 2
relaxFbβ 0.723 # 2
Salient Object Detection ECSSD U2-Net+ (RSU) Model Size (MB) 4.7 # 2
{max}Fβ 0.943 # 1
MAE 0.041 # 3
Fwβ 0.885 # 2
Sm 0.918 # 1
relaxFbβ 0.808 # 2
Saliency Detection HKU-IS U2-Net+ MAE 0.037 # 3
{max}Fβ 0.928 # 1
Fwβ 0.867 # 1
Sm 0.908 # 1
relaxFbβ 0.794 # 1
Salient Object Detection HKU-IS U2-Net (Ours) MAE 0.031 # 1
Salient Object Detection PASCAL-S U2-Net+ {max}Fβ 0.849 # 1
MAE 0.086 # 2
Fwβ 0.768 # 1
Sm 0.831 # 1
relaxFbβ 0.627 # 1
Salient Object Detection SOD U2-Net+ {max}Fβ 0.841 # 1
MAE 0.124 # 1
Fwβ 0.697 # 1
Sm 0.759 # 1
relaxFbβ 0.559 # 1

Methods