Syn-UI-Final-samples
收藏资源简介:
Syn-UI-Final Method Samples是一个小型视觉预览数据集,旨在为研究人员提供Syn-UI-Final生成方法的快速定性比较和检查。数据集包含200张合成水下图像,由四种不同的生成方法各生成50张图像:SyreaNet、WaterGAN、UWNR和CUT。对于每种方法,图像从训练分割和测试分割中各采样25张,使用固定种子(20260629)以确保可重复性。数据以分层目录结构组织,在images/目录下按方法名称(如syreanet/、watergan/等)和分割(train/、test/)存储图像文件,并附有一个metadata.csv文件。SyreaNet是一种基于深度感知的物理/规则水下合成方法;WaterGAN是一种基于生成对抗网络(GAN)的方法,利用深度和水域统计信息进行水下图像生成;UWNR使用预训练网络,结合干净图像、伪深度和水参考信息生成水下图像;CUT则是一种未配对的图像到图像翻译方法,将ImageNet风格的自然图像转换为Real-U风格的水下图像。数据集仅包含生成的预览图像,不包含原始服务器路径、账户凭证或完整数据集清单,主要用于研究交流、视觉评估和合成方法比较。用户在使用或重新分发完整衍生数据集前,应检查原始数据源的许可证和使用条款。该数据集适用于图像到图像任务、水下图像合成、合成数据生成和计算机视觉研究。
Syn-UI-Final Method Samples is a small-scale visual preview dataset designed to provide researchers with rapid qualitative comparison and inspection of the Syn-UI-Final generation methods. The dataset contains 200 synthesized underwater images, with 50 images generated by each of the four distinct generation methods: SyreaNet, WaterGAN, UWNR, and CUT. For each method, 25 images are sampled from both the training split and test split, using a fixed random seed (20260629) to ensure reproducibility. The data is organized in a hierarchical directory structure: image files are stored under the "images/" directory, categorized by method names (e.g., syreanet/, watergan/, etc.) and data splits (train/, test/), accompanied by a metadata.csv file. SyreaNet is a depth-aware physical/rule-based underwater image synthesis method; WaterGAN is a generative adversarial network (GAN)-based method that leverages depth and water body statistical information for underwater image generation; UWNR uses a pre-trained network to generate underwater images by combining clean images, pseudo-depth maps, and water reference information; CUT is an unpaired image-to-image translation method that converts ImageNet-style natural images into Real-U-style underwater images. This dataset only contains the generated preview images, and does not include original server paths, account credentials, or the full dataset inventory. It is primarily intended for research exchange, visual evaluation, and comparison of synthesis methods. Users should review the licenses and terms of use of the original data sources before using or redistributing the full derivative dataset. This dataset is applicable to image-to-image tasks, underwater image synthesis, synthetic data generation, and computer vision research.
数据集概述:Syn-UI-Final Method Samples
该数据集是 Syn-UI-Final 数据集的视觉预览样本,包含前四种生成方法产出的200张水下合成图像,用于快速视觉检查和质量比较,并非完整数据集。
- 许可协议:其他(需自行确认原数据源的许可与使用条款)
- 任务类别:图像到图像
- 标签:水下图像合成、合成数据、ImageNet、Real-U、WaterGAN、SyreaNet、UWNR、CUT
数据构成
| 生成方法 | 图像数量 | 训练集图像数 | 测试集图像数 |
|---|---|---|---|
| SyreaNet | 50 | 25 | 25 |
| WaterGAN | 50 | 25 | 25 |
| UWNR | 50 | 25 | 25 |
| CUT | 50 | 25 | 25 |
所有样本均使用固定随机种子(20260629)从每个方法的训练集和测试集中各抽取25张。
目录结构
text images/ syreanet/ train/ (25张图像) test/ (25张图像) watergan/ train/ (25张图像) test/ (25张图像) uwnr/ train/ (25张图像) test/ (25张图像) cut/ train/ (25张图像) test/ (25张图像) metadata.csv
生成方法说明
- SyreaNet:基于深度信息的物理/规则驱动水下图像合成。
- WaterGAN:基于GAN的水下图像生成,利用深度与水域统计特征。
- UWNR:基于预训练网络的水下图像生成,使用干净图像、伪深度与水质参考信息。
- CUT:无监督图像到图像翻译,将 ImageNet 风格的自然图像转换为 Real-U 风格的水下图像。
注意事项
- 数据集中仅包含生成的预览图像,不包含原始服务器路径、账号信息或完整数据集清单。
- 该样本仅供研究交流与定性比较,在分发或使用完整衍生数据集前,请核查原数据源的许可与使用条款。




