Learning Enriched Features via Selective State Spaces Model for Efficient Image Deblurring

Learning Enriched Features via Selective State Spaces Model for Efficient Image Deblurring

5 Apr 2024 | Hu Gao, Depeng Dang
The paper introduces ALGNet, an efficient image deblurring network that leverages the selective state spaces model (SSM) to aggregate enriched and accurate features. The network consists of multiple ALGBlocks, each containing a capturing local and global features module (CLGF) and a feature aggregation module (FA). The CLGF module captures long-range dependency features using SSM and models local connectivity with simplified channel attention, addressing issues like local pixel forgetting and channel redundancy. The FA module emphasizes the importance of local features by dynamically recalibrating weights during the aggregation process. Experimental results on various benchmarks demonstrate that ALGNet outperforms state-of-the-art methods, achieving superior performance while maintaining computational efficiency.The paper introduces ALGNet, an efficient image deblurring network that leverages the selective state spaces model (SSM) to aggregate enriched and accurate features. The network consists of multiple ALGBlocks, each containing a capturing local and global features module (CLGF) and a feature aggregation module (FA). The CLGF module captures long-range dependency features using SSM and models local connectivity with simplified channel attention, addressing issues like local pixel forgetting and channel redundancy. The FA module emphasizes the importance of local features by dynamically recalibrating weights during the aggregation process. Experimental results on various benchmarks demonstrate that ALGNet outperforms state-of-the-art methods, achieving superior performance while maintaining computational efficiency.
Reach us at info@study.space
[slides and audio] Learning Enriched Features via Selective State Spaces Model for Efficient Image Deblurring