Scholarlink

In recent years, intelligent segmentation and element combination of clothing patterns have become important research directions in the field of computer vision and industrial design. Aiming at the shortcomings of traditional convolutional networks in boundary recognition and combination consistency, this paper proposes a MGNN-OEGC model based on multi-scale graph neural network and element energy optimization, which integrates pyramid pooling and conditional random field reasoning to improve segmentation accuracy and combination rationality. Experimental results show that the mIoU of this method reaches 88.2% on the LabCloth-Pattern dataset, the Boundary F1 score is improved to 86.8%, and the ECR and LCS reach 92.3% and 86.7% respectively, which are significantly better than the comparison algorithms such as GraphCut-Comb and AffinityNet-Align. In the high-difficulty sample of CustomCombination, MGNN-OEGC maintains 82.7% mIoU and 79.1 CFS, showing stable generalization ability. This method can provide new ideas and practical basis for the accurate analysis and high-quality combination of complex clothing patterns.

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

Semantic Segmentation and Combination Method of Clothing Pattern Elements Combined with Graph Neural Network

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Junnan Cai,
  • Senrong Lin,
  • Yilin Li,
  • Yilong Wu

摘要

In recent years, intelligent segmentation and element combination of clothing patterns have become important research directions in the field of computer vision and industrial design. Aiming at the shortcomings of traditional convolutional networks in boundary recognition and combination consistency, this paper proposes a MGNN-OEGC model based on multi-scale graph neural network and element energy optimization, which integrates pyramid pooling and conditional random field reasoning to improve segmentation accuracy and combination rationality. Experimental results show that the mIoU of this method reaches 88.2% on the LabCloth-Pattern dataset, the Boundary F1 score is improved to 86.8%, and the ECR and LCS reach 92.3% and 86.7% respectively, which are significantly better than the comparison algorithms such as GraphCut-Comb and AffinityNet-Align. In the high-difficulty sample of CustomCombination, MGNN-OEGC maintains 82.7% mIoU and 79.1 CFS, showing stable generalization ability. This method can provide new ideas and practical basis for the accurate analysis and high-quality combination of complex clothing patterns.