𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡丨Diffusion-based Image Dehazing with Unpaired Data

AI Researcher & Co-author

𝐌𝐨𝐭𝐢𝐯𝐚𝐭𝐢𝐨𝐧 & 𝐂𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: Synthetic hazy-clear pairs often fail to represent real haze. To tackle this limitation, we introduce Diff-Dehazer, which combines pretrained diffusion priors, cycle-consistent unpaired learning, physical guidance and text-aware conditioning.

𝐑𝐞𝐬𝐮𝐥𝐭𝐬 & 𝐕𝐞𝐧𝐮𝐞𝐬: Improved dehazing on multiple real-world datasets without paired training images. Published at AAAI 2025.

𝐋𝐢𝐧𝐤：https://ojs.aaai.org/index.php/AAAI/article/download/32469/34624