P‐5.12: Multi‐Scale Fusion Enhanced Sparse Transformer for Image Deraining
SID Symposium Digest of Technical Papers · 2026-08 · 57 (S1) · 1355-1358
DOI: 10.1002/sdtp.70003_358 ↗Kurz gefasst
Fachartikel in der Zeitschrift SID Symposium Digest of Technical Papers, online erschienen am 11.09.2026. Verfasst von Enfang Xu. Die Arbeit nennt 13 Literaturangaben.
Abstract (Originalsprache)
Single image deraining is an important problem in thefield of low‐ level vision, aiming to restore a clear background from an observed image degraded by rain streaks. Existing Transformer‐based models typically use the similarity of all tokens in the query‐key‐ value pair for feature aggregation; if the query and key tokens are irrelevant, the computed attention values can interfere with the restoration of the clear image. Moreover, standard feed‐forward networks struggle to effectively model multi‐scale information, which is crucial for image restoration. To address the above issues, this paper, based on the framework of the Sparse Transformer Deraining Network (DRSformer), proposes an Enhanced Mixed‐ Scale Feed‐Forward Network (MSFN+). This strategy explicitly enhances the model's ability to capture rain streak features at different scales by optimizing its Mixed‐Scale Feed‐forward Network (MSFN), further strengthening the model's multi‐scale feature extraction and fusion capabilities. Experimental results on the Rain200L benchmark dataset show that the proposed method achieves good performance compared to current methods.
Zitieren
APA
Xu, E. (2026). P‐5.12: Multi‐Scale Fusion Enhanced Sparse Transformer for Image Deraining. SID Symposium Digest of Technical Papers, 57(S1), 1355-1358. https://doi.org/10.1002/sdtp.70003_358
Vancouver
Xu E. P‐5.12: Multi‐Scale Fusion Enhanced Sparse Transformer for Image Deraining. SID Symposium Digest of Technical Papers. 2026;57(S1):1355-1358. doi:10.1002/sdtp.70003_358
BibTeX
@article{Xu2026,
author = {Xu, Enfang},
title = {P‐5.12: Multi‐Scale Fusion Enhanced Sparse Transformer for Image Deraining},
journal = {SID Symposium Digest of Technical Papers},
year = {2026},
volume = {57},
number = {S1},
pages = {1355--1358},
doi = {10.1002/sdtp.70003_358}
}
EUVIMED