moondream/brackish_underwater

这是一个关于温带咸淡水环境中水下镜头的目标检测数据集，包含多种海洋动物。数据集由奥尔堡大学的Pedersen、Haurum、Gade和Moeslund在CVPR 2019研讨会上提出，图像来自长期海洋物种监测的固定摄像头。这是首批来自温带水域的标注水下图像数据集之一。数据集包含6种类别：螃蟹、鱼类、水母、虾、小鱼和海星。数据集分为训练集（9,967张图像）、测试集（1,238张图像）和验证集（1,239张图像）。每个示例包含一张1920x1080的图像和检测到的对象列表，每个对象包含类别名称和归一化坐标的边界框。

An object detection dataset of underwater footage from brackish water environments in temperate waters, featuring various marine animals. The dataset was introduced by Pedersen, Haurum, Gade, and Moeslund from Aalborg University at the CVPR 2019 workshop. The images were captured from permanently mounted cameras for long-term marine species monitoring. It is one of the first annotated underwater image datasets from temperate waters. The dataset includes 6 classes: crab, fish, jellyfish, shrimp, small_fish, and starfish. It is split into train (9,967 images), test (1,238 images), and valid (1,239 images) sets. Each example contains a 1920x1080 image and a list of detected objects, each with a class name and bounding boxes in normalized coordinates.