This paper introduces the Underwater Image Enhancement Benchmark (UIEB), a large-scale real-world dataset containing 950 underwater images, 890 of which have corresponding reference images. The dataset includes 60 challenging images without satisfactory reference images. The UIEB is used to evaluate state-of-the-art underwater image enhancement algorithms both qualitatively and quantitatively. A CNN-based model called Water-Net is proposed and trained on the UIEB to demonstrate its generalization for training CNNs. The benchmark evaluations and Water-Net demonstrate the performance and limitations of current algorithms, highlighting the need for further research. The dataset and code are available at https://li-chongyi.github.io/proj_benchmark.html. The UIEB provides a platform for evaluating underwater image enhancement algorithms and enables the training of CNNs for underwater image enhancement. The paper also discusses existing methods, evaluation metrics, and datasets for underwater image enhancement. It highlights the challenges of underwater image enhancement due to complex environments and lighting conditions, and the limitations of current methods. The proposed Water-Net model is evaluated on the UIEB and shows promising results in enhancing underwater images. The paper concludes that the UIEB is a valuable resource for advancing underwater image enhancement research.This paper introduces the Underwater Image Enhancement Benchmark (UIEB), a large-scale real-world dataset containing 950 underwater images, 890 of which have corresponding reference images. The dataset includes 60 challenging images without satisfactory reference images. The UIEB is used to evaluate state-of-the-art underwater image enhancement algorithms both qualitatively and quantitatively. A CNN-based model called Water-Net is proposed and trained on the UIEB to demonstrate its generalization for training CNNs. The benchmark evaluations and Water-Net demonstrate the performance and limitations of current algorithms, highlighting the need for further research. The dataset and code are available at https://li-chongyi.github.io/proj_benchmark.html. The UIEB provides a platform for evaluating underwater image enhancement algorithms and enables the training of CNNs for underwater image enhancement. The paper also discusses existing methods, evaluation metrics, and datasets for underwater image enhancement. It highlights the challenges of underwater image enhancement due to complex environments and lighting conditions, and the limitations of current methods. The proposed Water-Net model is evaluated on the UIEB and shows promising results in enhancing underwater images. The paper concludes that the UIEB is a valuable resource for advancing underwater image enhancement research.