3-D Convolutional Neural Networks for RGB-D Salient Object Detection and Beyond

Area of Science:

• Computer Vision
• Deep Learning
• Artificial Intelligence

Background:

• RGB-depth salient object detection (SOD) is a growing research area.
• Existing deep learning models often struggle with effective cross-modal feature fusion.

Purpose of the Study:

• To propose a novel 3-D convolutional neural network (CNN) for RGB-D SOD.
• To enhance the integration of RGB and depth information for improved salient object detection.

Main Methods:

• Introduced RD3D, a model utilizing 3-D CNNs for prefusion in the encoder and in-depth fusion in the decoder.
• Developed RD3D+ by disentangling 3-D convolutions into spatial and temporal components, optimizing efficiency.
• Incorporated channel-modality attention mechanisms to refine feature representation.

Main Results:

• RD3D and RD3D+ demonstrated superior performance compared to 14 state-of-the-art RGB-D SOD methods.
• The proposed progressive-fusion strategy effectively integrated RGB and depth modalities.
• Experiments on seven benchmark datasets validated the model's effectiveness across key evaluation metrics.

Conclusions:

• The novel 3-D CNN approach significantly advances RGB-D salient object detection.
• RD3D and RD3D+ offer a more effective and efficient solution for cross-modal feature fusion in SOD.
• The study provides a valuable contribution to the field of computer vision and deep learning for image analysis.