GCoNet+: A Stronger Group Collaborative Co-Salient Object Detector.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Co-salient object detection (CoSOD) aims to identify multiple related objects in an image.
• Existing methods struggle with accurately detecting co-salient objects in complex natural scenes.

Purpose of the Study:

• To introduce GCoNet+, a novel end-to-end group collaborative learning network for efficient and effective CoSOD.
• To achieve state-of-the-art performance in CoSOD by mining consensus representations.

Main Methods:

• GCoNet+ employs a group affinity module (GAM) for intra-group compactness and a group collaborating module (GCM) for inter-group separability.
• Incorporates a recurrent auxiliary classification module (RACM), confidence enhancement module (CEM), and group-based symmetric triplet (GST) loss.
• Achieves high efficiency with a processing speed of 250 frames per second.

Main Results:

• GCoNet+ significantly outperforms 12 existing state-of-the-art models on CoCA, CoSOD3k, and CoSal2015 benchmarks.
• Demonstrates superior accuracy and efficiency in co-salient object detection tasks.
• The proposed modules (GAM, GCM, RACM, CEM, GST) contribute to improved detection performance.

Conclusions:

• GCoNet+ represents a significant advancement in co-salient object detection.
• The network's novel approach to mining consensus representations enhances detection accuracy and efficiency.
• The open-sourced code facilitates further research and development in the field.