[论文解读] FFA-Net: Feature Fusion Attention Network for Single Image Dehazing

Qin Xu, Zhilin Wang+3 more|arXiv (Cornell University)|Nov 18, 2019

Image Enhancement Techniques参考文献 28被引用 120

一句话总结

FFA-Net 引入一个特征注意力模块，将通道注意力与像素注意力相结合，并结合局部残差学习与自适应特征融合，以显著提升单图像去雾性能。它在 RESIDE SOTS 数据集上实现了最先进的结果。

ABSTRACT

In this paper, we propose an end-to-end feature fusion at-tention network (FFA-Net) to directly restore the haze-free image. The FFA-Net architecture consists of three key components: 1) A novel Feature Attention (FA) module combines Channel Attention with Pixel Attention mechanism, considering that different channel-wise features contain totally different weighted information and haze distribution is uneven on the different image pixels. FA treats different features and pixels unequally, which provides additional flexibility in dealing with different types of information, expanding the representational ability of CNNs. 2) A basic block structure consists of Local Residual Learning and Feature Attention, Local Residual Learning allowing the less important information such as thin haze region or low-frequency to be bypassed through multiple local residual connections, let main network architecture focus on more effective information. 3) An Attention-based different levels Feature Fusion (FFA) structure, the feature weights are adaptively learned from the Feature Attention (FA) module, giving more weight to important features. This structure can also retain the information of shallow layers and pass it into deep layers. The experimental results demonstrate that our proposed FFA-Net surpasses previous state-of-the-art single image dehazing methods by a very large margin both quantitatively and qualitatively, boosting the best published PSNR metric from 30.23db to 36.39db on the SOTS indoor test dataset. Code has been made available at GitHub.

研究动机与目标

• Motivate robust single image dehazing under uneven haze distribution.
• Develop an end-to-end network that selectively focuses on informative features and pixels.
• Incorporate multi-level feature fusion while preserving shallow information.
• Demonstrate substantial performance gains over prior dehazing methods on RESIDE SOTS.

提出的方法

• Introduce the Feature Attention (FA) module that combines Channel Attention and Pixel Attention.
• Build a basic block with Local Residual Learning and FA to bypass less important information.
• Use an Attention-based Feature Fusion (FFA) structure to adaptively weight and fuse features from multiple group architectures.
• Assemble Group Architectures with skip connections and a final reconstruction with global residual learning.
• Optimize with L1 loss for image restoration quality.
• Train on RESIDE SOTS indoor/outdoor with data augmentation and Adam optimizer.

实验结果

研究问题

• RQ1 Can a combined Channel and Pixel Attention module effectively handle uneven haze distribution and channel-wise information in hazy images?
• RQ2 Does local residual learning improve training stability and preserve important low-frequency/transparent haze information?
• RQ3 Can adaptive feature fusion across multi-level features preserve shallow details while enhancing deep representations to outperform prior dehazing methods?

主要发现

• FFA-Net outperforms prior methods on RESIDE SOTS with PSNR and SSIM gains (Indoor PSNR 36.39 dB, SSIM 0.9886; Outdoor PSNR 33.57 dB, SSIM 0.9840).
• The FA module (Channel + Pixel Attention) provides adaptive weights that emphasize thick haze regions and high-frequency textures.
• Local Residual Learning stabilizes training and improves performance.
• Attention-based Feature Fusion preserves shallow information and enables effective deep feature integration.
• Ablation shows each component (FA, LRL, and FFA) contributes to improved PSNR, with full configuration achieving 35.77 dB in a reduced setting and 36.39 dB in the reported setup.