CIR-Net: Cross-Modality Interaction and Refinement for RGB-D Salient Object Detection.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Salient Object Detection (SOD) aims to identify visually distinct objects.
• RGB-D data, combining color and depth information, offers richer context for SOD.
• Effective fusion of cross-modality information remains a challenge in RGB-D SOD.

Purpose of the Study:

• To propose a novel Convolutional Neural Network (CNN) model, CIR-Net, for RGB-D salient object detection.
• To enhance the capture and utilization of cross-modality information through interaction and refinement mechanisms.
• To achieve state-of-the-art performance in RGB-D SOD.

Main Methods:

• Introduced a progressive attention-guided integration unit for feature fusion in the encoder stage.
• Developed a convergence aggregation structure with an importance-gated fusion unit in the decoder stage.
• Implemented a refinement middleware with self-modality and cross-modality attention refinement units.

Main Results:

• CIR-Net demonstrates superior performance compared to existing methods on six benchmark datasets.
• Qualitative and quantitative experiments validate the effectiveness of the proposed cross-modality interaction and refinement.
• The model successfully refines features through successive attention mechanisms.

Conclusions:

• The proposed CIR-Net effectively leverages cross-modality information for RGB-D salient object detection.
• The novel interaction and refinement strategies significantly improve SOD performance.
• CIR-Net represents a significant advancement in the field of RGB-D salient object detection.