GE9X/FishDet-M
收藏资源简介:
FishDet-M是一个用于多样化水下环境中鱼类检测的最大统一基准数据集,旨在推动生态监测、水产养殖自动化和海洋机器人技术的发展。数据集包含105,556张图像和296,885个标注的鱼类实例,涵盖了海洋、半咸水、水族馆和遮挡场景等多种环境。数据集采用COCO风格的标注格式,并评估了28种模型,包括YOLOv8-v12、DETR、R-CNN变体等。此外,数据集还利用CLIP-Based Selection技术进行自动模型选择。数据集整合了13个公共数据集,涵盖了珊瑚礁、开放水域、养殖池和池塘等多种场景,图像分辨率从78×53到4608×3456像素不等。
FishDet-M is the largest unified benchmark dataset for fish detection in diverse underwater environments, designed to advance ecological monitoring, aquaculture automation, and marine robotics. The dataset contains 105,556 images with 296,885 annotated fish instances, covering marine, brackish, aquarium, and occluded scenes. It features COCO-style annotations with bounding boxes and evaluates 28 models including YOLOv8-v12, DETR, R-CNN variants, and more. Additionally, the dataset employs CLIP-Based Selection for automatic model selection. FishDet-M integrates 13 public datasets spanning coral reefs, open water, aquaculture tanks, and ponds, with image resolutions ranging from 78×53 to 4608×3456 pixels.




