GBU-UCOD
收藏资源简介:
GBU-UCOD是由哈尔滨工程大学与大湾区大学联合创建的首个高分辨率(2K)水下伪装物体检测基准数据集,专注于海洋垂直分带研究。该数据集包含1,052组图像-掩码对,覆盖从表层(0-200米)至超深渊带(4000米+)的典型生物样本,尤其包含大量透明生物和脆弱拓扑结构的深海生物标注。数据通过严格像素级标注策略采集,重点保障了纤细肢体(如触须)的拓扑连贯性。该数据集旨在解决现有浅水数据集的深度偏差问题,为深海光学衰减环境和复杂生物形态的算法研究提供关键数据支持。
GBU-UCOD is the first high-resolution (2K) benchmark dataset for underwater camouflage object detection, co-developed by Harbin Engineering University and Greater Bay Area University, focusing on marine vertical zonation research. This dataset contains 1,052 sets of image-mask pairs, covering typical biological samples from the surface layer (0-200 meters) to the hadal zone (4000 meters and above), and includes a large number of annotations for transparent organisms and deep-sea creatures with fragile topological structures. The dataset was collected with strict pixel-level annotation strategies, with particular emphasis on ensuring the topological coherence of slender appendages such as tentacles. This dataset aims to address the depth bias issue of existing shallow-water datasets, and provides critical data support for algorithm research on deep-sea optical attenuation environments and complex biological morphologies.
GBU-UCOD数据集概述
数据集名称
GBU-UCOD
数据集来源
- 代码仓库地址:https://github.com/Wuwenji18/GBU-UCOD
数据集描述
根据提供的README文件内容,该数据集未提供进一步的详细描述、背景、数据内容、格式、规模或使用方式等信息。




