Single UHD Image Dehazing Via Interpretable Pyramid Network

Currently, most single image dehazing models cannot run an ultra-high-resolution (UHD) with a GPU shader in real-time. To address the problem, we introduce principle of infinite approximation Taylor's theorem Laplace pyramid pattern to build model which is capable handling 4K hazy images The N branch networks network correspond constraint terms theorem. Low-order polynomials reconstruct low-frequency information (e.g. color, illumination). High-order regress high-frequency texture). In addition, propose Tucker reconstruction-based regularization term that acts on each model. It further constrains generation anomalous signals feature space. Extensive experimental results demonstrate our approach can not only haze real-time (80FPS) but also has unparalleled interpretability. developed method achieves state-of-the-art (SOTA) performance two benchmarks (O/I-HAZE) and updated 4KID dataset while providing reliable groundwork for subsequent optimization schemes.