Densely Deformable Efficient Salient Object Detection Network

TL;DR

This paper introduces a Densely Deformable Network (DDNet) for efficient Salient Object Detection that leverages deformable convolutions for improved accuracy and generalization, validated on multiple datasets.

Contribution

The paper proposes a novel DDNet architecture utilizing deformable convolutions and transposed convolutions for enhanced SOD performance and generalization, along with a new dataset for diverse scenario testing.

Findings

Outperforms 22 existing methods in efficiency and effectiveness

Demonstrates better generalization on a new cross-dataset (S-SOD)

Provides publicly available code and dataset for further research

Abstract

Salient Object Detection (SOD) domain using RGB-D data has lately emerged with some current models' adequately precise results. However, they have restrained generalization abilities and intensive computational complexity. In this paper, inspired by the best background/foreground separation abilities of deformable convolutions, we employ them in our Densely Deformable Network (DDNet) to achieve efficient SOD. The salient regions from densely deformable convolutions are further refined using transposed convolutions to optimally generate the saliency maps. Quantitative and qualitative evaluations using the recent SOD dataset against 22 competing techniques show our method's efficiency and effectiveness. We also offer evaluation using our own created cross-dataset, surveillance-SOD (S-SOD), to check the trained models' validity in terms of their applicability in diverse scenarios. The results indicate that the current models have limited generalization potentials, demanding further research in this direction. Our code and new dataset will be publicly available at https://github.com/tanveer-hussain/EfficientSOD