Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CPDR: Towards Highly-Efficient Salient Object Detection via Crossed Post-decoder Refinement

Yijie Li, Hewei Wang, Aggelos Katsaggelos

TL;DR

The paper tackles the challenge of efficient salient object detection by introducing Crossed Post-decoder Refinement (CPDR), a lightweight refinement module that sits after standard decoders in FPN/U-Net architectures. CPDR comprises three components: Attention Down-Sample Fusion (ADF), Attention Up-Sample Fusion (AUF), and Dual Attention Cross Fusion (DACF), which enable cross-level, cross-attention feature merging while keeping computational demands low. Training uses DICE and IoU losses with deep supervision, achieving 0.041 MAE using only 1.66M parameters and outperforming prior SOTA on five benchmarks. The approach offers a practical plug-and-play solution for resource-constrained deployment and shows potential for extending to other computer vision tasks beyond SOD.

Abstract

Most of the current salient object detection approaches use deeper networks with large backbones to produce more accurate predictions, which results in a significant increase in computational complexity. A great number of network designs follow the pure UNet and Feature Pyramid Network (FPN) architecture which has limited feature extraction and aggregation ability which motivated us to design a lightweight post-decoder refinement module, the crossed post-decoder refinement (CPDR) to enhance the feature representation of a standard FPN or U-Net framework. Specifically, we introduce the Attention Down Sample Fusion (ADF), which employs channel attention mechanisms with attention maps generated by high-level representation to refine the low-level features, and Attention Up Sample Fusion (AUF), leveraging the low-level information to guide the high-level features through spatial attention. Additionally, we proposed the Dual Attention Cross Fusion (DACF) upon ADFs and AUFs, which reduces the number of parameters while maintaining the performance. Experiments on five benchmark datasets demonstrate that our method outperforms previous state-of-the-art approaches.

CPDR: Towards Highly-Efficient Salient Object Detection via Crossed Post-decoder Refinement

TL;DR

The paper tackles the challenge of efficient salient object detection by introducing Crossed Post-decoder Refinement (CPDR), a lightweight refinement module that sits after standard decoders in FPN/U-Net architectures. CPDR comprises three components: Attention Down-Sample Fusion (ADF), Attention Up-Sample Fusion (AUF), and Dual Attention Cross Fusion (DACF), which enable cross-level, cross-attention feature merging while keeping computational demands low. Training uses DICE and IoU losses with deep supervision, achieving 0.041 MAE using only 1.66M parameters and outperforming prior SOTA on five benchmarks. The approach offers a practical plug-and-play solution for resource-constrained deployment and shows potential for extending to other computer vision tasks beyond SOD.

Abstract

Most of the current salient object detection approaches use deeper networks with large backbones to produce more accurate predictions, which results in a significant increase in computational complexity. A great number of network designs follow the pure UNet and Feature Pyramid Network (FPN) architecture which has limited feature extraction and aggregation ability which motivated us to design a lightweight post-decoder refinement module, the crossed post-decoder refinement (CPDR) to enhance the feature representation of a standard FPN or U-Net framework. Specifically, we introduce the Attention Down Sample Fusion (ADF), which employs channel attention mechanisms with attention maps generated by high-level representation to refine the low-level features, and Attention Up Sample Fusion (AUF), leveraging the low-level information to guide the high-level features through spatial attention. Additionally, we proposed the Dual Attention Cross Fusion (DACF) upon ADFs and AUFs, which reduces the number of parameters while maintaining the performance. Experiments on five benchmark datasets demonstrate that our method outperforms previous state-of-the-art approaches.
Paper Structure (16 sections, 8 equations, 4 figures, 2 tables)

This paper contains 16 sections, 8 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Crossed Post-decoder Refinement (CPDR)
  4. Overview of CPDR
  5. Attention Down-Sample Fusion (ADF)
  6. Attention Up-Sample Fusion (AUF)
  7. Dual Attention Cross Fusion (DACF)
  8. Objective Functions
  9. Experiments
  10. Datasets
  11. Evaluation Metrics
  12. Implementation Detail
  13. Ablation Study
  14. Qualitative Evaluation
  15. Quantitative Evaluation
  16. ...and 1 more sections

Figures (4)

  • Figure 1: Overall Pipeline of CPDR
  • Figure 2: Demonstration of CPDR Modules
  • Figure 3: Qualitative comparison on DUTS-TE dataset between our methods and previous lightweight state-of-the-art approaches
  • Figure 4: Illustration of F-measure curves on five benchmark datasets