Scholarlink

<p>In the realm of computer vision, accurately detecting camouflaged objects amidst highly similar backgrounds remains a formidable challenge. Existing methods for collaborative camouflaged object detection (CoCOD) often neglect the co-representation information within camouflaged object groups. This paper introduces the Consensus-Aware Foreground Refinement Network (CFNet), a novel framework designed to address this issue. CFNet comprises a Consensus Awareness Module (CAM) to capture shared semantic cues across images, a Foreground Feature Refinement Module (FRM) to enhance foreground representations, and a Foreground-induced Integrity Aggregation Decoder (FIAD) to fuse collaborative and refined foreground features. Experimental results on the CoCOD8K dataset demonstrate that CFNet outperforms state-of-the-art CoCOD, camouflaged object detection (COD), and collaborative salient object detection (CoSOD) models, achieving impressive performance with an F-measure of 0.750 and an E-measure of 0.896. This work highlights the potential of CFNet in enhancing the accuracy of collaborative camouflaged object detection in complex environments. The code can be obtained at <a href="https://github.com/Firefly-m/CFNet">https://github.com/Firefly-m/CFNet</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Consensus aware foreground refinement network for collaborative camouflaged object detection

  • 期刊论文

  • 发表日期：

    2025年9月7日

查看全文
  • Hongbo Bi,
  • Disen Mo,
  • Shiyuan Li,
  • Cong Zhang

摘要

In the realm of computer vision, accurately detecting camouflaged objects amidst highly similar backgrounds remains a formidable challenge. Existing methods for collaborative camouflaged object detection (CoCOD) often neglect the co-representation information within camouflaged object groups. This paper introduces the Consensus-Aware Foreground Refinement Network (CFNet), a novel framework designed to address this issue. CFNet comprises a Consensus Awareness Module (CAM) to capture shared semantic cues across images, a Foreground Feature Refinement Module (FRM) to enhance foreground representations, and a Foreground-induced Integrity Aggregation Decoder (FIAD) to fuse collaborative and refined foreground features. Experimental results on the CoCOD8K dataset demonstrate that CFNet outperforms state-of-the-art CoCOD, camouflaged object detection (COD), and collaborative salient object detection (CoSOD) models, achieving impressive performance with an F-measure of 0.750 and an E-measure of 0.896. This work highlights the potential of CFNet in enhancing the accuracy of collaborative camouflaged object detection in complex environments. The code can be obtained at https://github.com/Firefly-m/CFNet.