<p>Existing Few-Shot Semantic Segmentation (FSS) methods often focus on extracting semantic information from support images to guide the segmentation of query images, while less attention is paid to exploring the query branch. However, due to the limited number of support images, there exists significant intra-class variance between support and query images. Additionally, relying solely on a single support prototype to guide query segmentation often leads to inaccurate segmentation boundaries in the prediction results, which can affect the model’s performance. In this study, we simultaneously consider the information extraction from both the support and query branches and propose a Dual-branch Aggregation and Edge Refinement (DAER) network for accurate query image segmentation. Specifically, to better explore the information from the query branch, we introduce an Initial Mask Generation Module (IMGM) that generates an initial mask for the query image. Furthermore, we propose a Dual-Branch Aggregation Module (DBAM) that simultaneously captures information from both the support and query branches. Finally, an Edge Refinement Module (ERM) is introduced to integrate more query-specific positional information into the network. Extensive experiments on standard few-shot semantic segmentation benchmarks, including PASCAL-<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="530_2025_1718_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="16" /> </InlineMediaObject> <EquationSource Format="TEX">\(5^i\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mn>5</mn> <mi>i</mi> </msup> </math></EquationSource> </InlineEquation> and COCO-<InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="530_2025_1718_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="23" /> </InlineMediaObject> <EquationSource Format="TEX">\(20^i\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mn>20</mn> <mi>i</mi> </msup> </math></EquationSource> </InlineEquation>, demonstrate the effectiveness of our proposed method.</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

Dual-branch aggregation and edge refinement network for few shot semantic segmentation

  • Qingsong Tang,
  • Yalei Ren,
  • Zhanghui Shan,
  • Chenyang Bao,
  • Yang Liu

摘要

Existing Few-Shot Semantic Segmentation (FSS) methods often focus on extracting semantic information from support images to guide the segmentation of query images, while less attention is paid to exploring the query branch. However, due to the limited number of support images, there exists significant intra-class variance between support and query images. Additionally, relying solely on a single support prototype to guide query segmentation often leads to inaccurate segmentation boundaries in the prediction results, which can affect the model’s performance. In this study, we simultaneously consider the information extraction from both the support and query branches and propose a Dual-branch Aggregation and Edge Refinement (DAER) network for accurate query image segmentation. Specifically, to better explore the information from the query branch, we introduce an Initial Mask Generation Module (IMGM) that generates an initial mask for the query image. Furthermore, we propose a Dual-Branch Aggregation Module (DBAM) that simultaneously captures information from both the support and query branches. Finally, an Edge Refinement Module (ERM) is introduced to integrate more query-specific positional information into the network. Extensive experiments on standard few-shot semantic segmentation benchmarks, including PASCAL- \(5^i\) 5 i and COCO- \(20^i\) 20 i , demonstrate the effectiveness of our proposed method.