
From Abyssal Darkness to Blinding Glare: A Benchmark on Extreme Exposure Correction in Real World
This paper introduces Real-world Extreme Exposure Dataset (REED) to improve extreme exposure correction in real world scenarios. The method is based on burst capturing with a range of exposures and accurate SIFT-based image alignment. The paper also introduces a method (CLIER) for extreme exposure correction based on luminance normalization, semantic awareness, diffusion, and iterative refinement. The experiments validate the efficacy of the proposed method.