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

In this paper, we design a simple yet powerful deep network architecture,U$^2$-Net, for salient object detection (SOD). The architecture of ourU$^2$-Net is a two-level nested U-structure. The design has the followingadvantages: (1) it is able to capture more contextual information fromdifferent scales thanks to the mixture of receptive fields of different sizesin our proposed ReSidual U-blocks (RSU), (2) it increases the depth of thewhole architecture without significantly increasing the computational costbecause of the pooling operations used in these RSU blocks. This architectureenables us to train a deep network from scratch without using backbones fromimage classification tasks. We instantiate two models of the proposedarchitecture, U$^2$-Net (176.3 MB, 30 FPS on GTX 1080Ti GPU) andU$^2$-Net$^{\dagger}$ (4.7 MB, 40 FPS), to facilitate the usage in differentenvironments. Both models achieve competitive performance on six SOD datasets.The code is available: https://github.com/NathanUA/U-2-Net.