[CVPR2025] VCP: Visual Consensus Prompting for Co-Salient Object Detection

🔥🔥🔥Our work has been accepted at CVPR 2025, and we will release the source code soon.

Framework Overview

Results

The prediction results of our model are available on google-drive|BaiduYun(fetch code: 0825)

Quantitative and qualitative comparison with SOTA methods

Some alternative schemes of VCP exhibit more efficient parameters and competitive performance

Extention to RGB-D CoSOD task

We use the most straightforward early fusion strategy, which does not introduce additional parameters, to validate the effectiveness and generalization of the proposed VCP for the RGB-D CoSOD task. Quantitative and qualitative comparison with SOTA methods：