DCD-Net: Weakly supervised decomposition learning for real-world image dehazing

Yi Huang, Yu Dong, Jiancheng Huang, Mingfu Yan, Shifeng Chen

Abstract: Highlights•Real-world hazy images differ greatly from synthetic ones used in model training.•Model performance degrades significantly in complex real-world environments.•Enhancing model performance in complex real-world scenarios is a key challenge.•Learning haze characteristics from real data is a valuable yet difficult research.