Scholarlink

In salient object detection, the importance of model lightweighting is increasingly highlighted, particularly in resource-constrained environments. To address this, we present SDNet, an efficient decoder that significantly reduces the number of decoder parameters while maintaining competitive model performance. Integrated with MobileNetv2, SDNet requires only an additional 60k parameters, demonstrating substantial potential in simplifying model complexity. Additionally, our research finds that intermediate predictions can also serve as an effective localization prior, guiding shallow features in the segmentation of salient objects. Building on this insight, we further designed a composite model, validating the feasibility of using coarse predictions to guide different models. This opens new research perspectives for salient object detection. Extensive experiments on five public datasets have demonstrated the effectiveness of our method. The source code will be released at https://github.com/YinYinOvO/SDNet .

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

SDNet: A Simple and Efficient Salient Object Detection Decoder with Only 60K Parameters

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Chiheng Zhou,
  • Zhengkai Wang,
  • Yongxia Zhou,
  • Chen Pan

摘要

In salient object detection, the importance of model lightweighting is increasingly highlighted, particularly in resource-constrained environments. To address this, we present SDNet, an efficient decoder that significantly reduces the number of decoder parameters while maintaining competitive model performance. Integrated with MobileNetv2, SDNet requires only an additional 60k parameters, demonstrating substantial potential in simplifying model complexity. Additionally, our research finds that intermediate predictions can also serve as an effective localization prior, guiding shallow features in the segmentation of salient objects. Building on this insight, we further designed a composite model, validating the feasibility of using coarse predictions to guide different models. This opens new research perspectives for salient object detection. Extensive experiments on five public datasets have demonstrated the effectiveness of our method. The source code will be released at https://github.com/YinYinOvO/SDNet .