“Underwater computer vision, as an important technological means for marine scientific research and marine engineering applications, has received widespread attention in recent years. However, compared to terrestrial environments, underwater imaging is significantly challenged by factors such as light attenuation, color shift, scattering effects, and equipment limitations. The construction of high-quality datasets is a key resource for promoting the development of underwater computer vision technology, and their quality and diversity directly affect the training effectiveness and performance of models. To comprehensively understand the development, advantages, and limitations of underwater datasets, this article provides a systematic review of the current major datasets, covering multiple key tasks such as underwater visual enhancement, underwater scene understanding, and underwater 3D reconstruction. In terms of underwater visual enhancement, datasets such as image/video enhancement, color correction and restoration, and super-resolution reconstruction were analyzed; In terms of understanding underwater scenes, the system has compiled representative datasets for tasks such as object classification, object detection, semantic segmentation, instance segmentation, saliency detection, disguised object detection, and object tracking; In terms of underwater 3D reconstruction, the development status of datasets related to techniques such as simultaneous localization and mapping (SLAM), neural radiation fields (NeRF), and three-dimensional Gaussian scattering (3D GS) was explored. Based on a comprehensive analysis of the construction methods, scale characteristics, and application scenarios of existing datasets, this article summarizes the main challenges currently faced and explores possible future development directions, providing comprehensive dataset resource references for relevant researchers and helping to promote in-depth research and widespread application of underwater computer vision technology.”