Learning real-world heterogeneous noise models with a benchmark dataset

Lu Sun, Jie Lin, Weisheng Dong, Xin Li, Jinjian Wu, Guangming Shi

Abstract: Highlights•Current realistic datasets could not describe diverse noise properties sufficiently.•Parametric noise models cannot characterize the real-world noise exactly.•Constructing a more comprehensive real-world benchmark dataset.•Developing a pixel-wise variance estimation network.•Improved real-world denoising and classification performance.