Bifurcated Backbone Strategy for RGB-D Salient Object Detection

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Multi-level feature fusion is crucial for object detection, segmentation, and classification across scales.
• Integrating multi-level features with multi-modal cues presents challenges in optimal aggregation and learning strategies.

Purpose of the Study:

• To propose a novel Bifurcated Backbone Strategy Network (BBS-Net) for RGB-D salient object detection.
• To address the challenges of feature aggregation and multi-modal learning in computer vision tasks.

Main Methods:

• Developed a Bifurcated Backbone Strategy (BBS) to regroup multi-level features into teacher and student features.
• Introduced a Depth-Enhanced Module (DEM) to extract informative depth cues from channel and spatial views.
• Implemented a complementary fusion strategy for RGB and depth modalities.

Main Results:

• BBS-Net significantly outperforms 18 state-of-the-art (SOTA) models on eight challenging datasets.
• Achieved approximately 4% improvement in S-measure compared to the top-ranked DMRA model.
• Demonstrated superior performance and generalization ability across various RGB-D datasets.

Conclusions:

• BBS-Net offers a simple, efficient, and backbone-independent architecture for RGB-D salient object detection.
• The proposed approach effectively leverages multi-level features and multi-modal cues.
• The study provides a powerful training set and publicly available resources for future research.