Learning Implicit Class Knowledge for RGB-D Co-Salient Object Detection With Transformers.

Area of Science:

• Computer Vision
• Artificial Intelligence

Background:

• Co-salient object detection using RGB-D data is challenging.
• Existing methods struggle with capturing complex co-occurring object patterns and global context.

Purpose of the Study:

• To propose an end-to-end transformer-based model for RGB-D co-salient object detection.
• To improve the exploitation of intra- and inter-saliency cues and complementary RGB-D information.

Main Methods:

• An end-to-end transformer model (CTNet) utilizing adaptive and common class tokens.
• Leveraging complementary RGB and depth map cues for enhanced token learning.
• Construction of the large-scale RGBD CoSal1k benchmark dataset.

Main Results:

• CTNet effectively captures intra- and inter-saliency cues.
• The model demonstrates superior performance on RGB-D co-salient object detection tasks.
• The proposed RGBD CoSal1k dataset facilitates robust model evaluation.

Conclusions:

• Transformer-based models with class tokens offer a promising direction for RGB-D co-salient object detection.
• CTNet provides an effective solution for segmenting co-occurring salient objects.
• The new dataset aids future research in this domain.