Diffusion models have recently achieved encouraging results in real-world image dehazing, according to a new arXiv submission. However, restoring high-resolution images remains challenging because haze is often spatially non-uniform and predictions can be inconsistent across local regions.
The new method, called Hihaze, employs a one-step diffusion process to address these difficulties. The abstract suggests that this approach targets the specific problems of uneven haze distribution and inconsistent local predictions, potentially enabling more practical high-resolution dehazing.
While the abstract provides limited details, the work highlights ongoing efforts to make diffusion-based dehazing viable for real-world, high-resolution imagery.