U-shaped Vision Mamba for Single Image Dehazing

U-shaped Vision Mamba for Single Image Dehazing

16 Feb 2024 | Zhuoran Zheng1 and Chen Wu2
The paper introduces U-shaped Vision Mamba (UVM-Net), an efficient single-image dehazing network designed to address the computational complexity and long-range dependency limitations of Transformer-based models. Inspired by State Space Sequence Models (SSMs), UVM-Net incorporates a Bi-SSM block that integrates local feature extraction with long-range dependency modeling. This design leverages the strengths of both convolutional layers and SSMs to enhance the performance of image dehazing tasks. Extensive experimental results demonstrate the effectiveness of UVM-Net, showing superior performance compared to state-of-the-art methods. The method is also evaluated on other image restoration tasks, such as low-light enhancement and deraining, further validating its broad applicability. The code for UVM-Net is available at <https://github.com/zxr-idam>.The paper introduces U-shaped Vision Mamba (UVM-Net), an efficient single-image dehazing network designed to address the computational complexity and long-range dependency limitations of Transformer-based models. Inspired by State Space Sequence Models (SSMs), UVM-Net incorporates a Bi-SSM block that integrates local feature extraction with long-range dependency modeling. This design leverages the strengths of both convolutional layers and SSMs to enhance the performance of image dehazing tasks. Extensive experimental results demonstrate the effectiveness of UVM-Net, showing superior performance compared to state-of-the-art methods. The method is also evaluated on other image restoration tasks, such as low-light enhancement and deraining, further validating its broad applicability. The code for UVM-Net is available at <https://github.com/zxr-idam>.
Reach us at info@study.space
[slides] U-shaped Vision Mamba for Single Image Dehazing | StudySpace