𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡丨Real-World Image Dehazing via Schrödinger Bridge

AI Researcher & Co-author

𝐌𝐨𝐭𝐢𝐯𝐚𝐭𝐢𝐨𝐧 & 𝐂𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: Real-world dehazing needs flexible distribution mapping without sacrificing image details, where we observe that schrödinger bridge serves as a promising solution. Therefore, we propose DehazeSB, which combines a Schrodinger Bridge with detail-preserving regularization and learned haze-aware CLIP guidance.

𝐑𝐞𝐬𝐮𝐥𝐭𝐬 & 𝐕𝐞𝐧𝐮𝐞𝐬: Improved real-world dehazing through direct hazy-to-clear transport in fewer sampling steps. Published at ICCV 2025.

𝐋𝐢𝐧𝐤：https://arxiv.org/pdf/2507.09524