Scholarlink

<p>Recent methods based on Implicit Neural Representation (INR) have shown great potential in arbitrary-scale super-resolution (ASSR), but they still face two key limitations: (1) static feature fusion in encoders lacks adaptability for multi-scale feature interactions, causing edge blurring and artifacts; (2) limited nonlinear modeling capacity in decoders hinders high-frequency detail recovery. To address these issues, we propose an ASSR algorithm based on multi-scale dynamic aggregation and gated cooperative decoding (MDGCN). Specifically, we design a Multi-scale Dynamic Aggregation Network (MDAN) as the encoder, which enhances feature extraction efficiency and accuracy through dynamic convolution layers, multi-branch parallel extraction blocks and cross-scale feature aggregation. Additionally, the Gated Cooperative Decoding Network (GCDN) serves as the decoder, improving nonlinear expressiveness and high-frequency detail recovery via dual interaction feature enhancement and gated cooperative decoding units. Experiments on five benchmark datasets show that our method achieves consistent and competitive improvements in both visual quality and quantitative metrics.</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

MDGCN: multi-scale dynamic aggregation and gated cooperative decoding for arbitrary-scale image super-resolution

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Huilin Liu,
  • Yu Sun,
  • Qiong Fang,
  • Jianyu Zhou,
  • Chenxi Hu,
  • Wanqi Ma,
  • Tao Wang

摘要

Recent methods based on Implicit Neural Representation (INR) have shown great potential in arbitrary-scale super-resolution (ASSR), but they still face two key limitations: (1) static feature fusion in encoders lacks adaptability for multi-scale feature interactions, causing edge blurring and artifacts; (2) limited nonlinear modeling capacity in decoders hinders high-frequency detail recovery. To address these issues, we propose an ASSR algorithm based on multi-scale dynamic aggregation and gated cooperative decoding (MDGCN). Specifically, we design a Multi-scale Dynamic Aggregation Network (MDAN) as the encoder, which enhances feature extraction efficiency and accuracy through dynamic convolution layers, multi-branch parallel extraction blocks and cross-scale feature aggregation. Additionally, the Gated Cooperative Decoding Network (GCDN) serves as the decoder, improving nonlinear expressiveness and high-frequency detail recovery via dual interaction feature enhancement and gated cooperative decoding units. Experiments on five benchmark datasets show that our method achieves consistent and competitive improvements in both visual quality and quantitative metrics.