iShip-1: 海面小目标船舶目标检测数据集

iShip-1: Maritime Small-Scale Ship Detection Dataset

数据集概览 Overview of the iShip-1 

iShip-1数据集一共有17236张多视角、多天气条件下拍摄的岸上和船上的图像和相应的候选框标注数据组成。iShip-1将船型按功能划分为5类，同时创新性地提供了小目标船的标注，为海面船舶检测提出全新的挑战。iShip-1提供VOC，COCO和YOLO格式的标注，以方便研究者们进行海面目标检测的算法研究，助力人工智能赋能智能海洋。

The iShip-1 dataset consists of a total of 17,236 images of shore and ship taken under multi-view and multi-weather conditions and the corresponding annotations. iShip-1 divides ship types into 5 categories, and also innovatively provides the labels of small-scale ships, which poses a brand-new challenge for ship detection. iShip-1 provides VOC, COCO, and YOLO formats to facilitate researchers to conduct algorithmic research on ship detection, and help AI to empower the smart ocean.

数据集介绍 Introduction to the iShip-1

我们重新标注了Seaships数据集[1]共7000张图像，并自行采集了10,236幅在不同天气条件下拍摄的岸上和船上图像。这些目标物按照功能划分为5类：散货船（Bulk Carrier），货船（Cargo Ship）, 客船（Passenger Ship）, 渔船（Fishing Vessel）, 娱乐用船（Pleasure Craft），而小目标船（在图片中所占面积比例小于0.5%）则统一被划分为其他船（Other Ship）。图像分辨率在800x600到6000x4000。

We re-labelled a total of 7,000 images from the Seaships dataset [1] and collected 10,236 shore and ship images taken under different weather conditions. These objects are classified into five categories according to their functions: Bulk Carrier, Cargo Ship, Passenger Ship, Fishing Vessel, and Pleasure Craft, while small objects (less than 0.5% of the area in the image) are uniformly classified as Other Ship. Image resolutions range from 800x600 to 6000x4000.

这些分类统计可视化结果如下：

The results of these categorical statistics are shown below:

我们使用多种模型在数据集上进行了基准测试，证明数据集在小目标检测算法领域是具有一定的研究价值的。

We benchmarked the dataset using a variety of models and proved that the dataset’s research value in the field of small object detection algorithms.

下图为yolov5在iShip-1上的PR曲线，其中小目标船检测对模型提出了更大的挑战。

The figure below shows the PR curve of yolov5 on iShip-1, where small-scale ship detection poses a greater challenge to the model.

部署推理视频【2】：

Result Video [2]:

数据集样例 Samples of the iShip-1

数据集使用说明  Instructions for use of the iShip-1

在images目录下存放了jpg格式的图片用于训练，labels目录下是YOLO格式的标注文件，annotations下是VOC格式的标注文件，coco目录下是COCO格式的标注文件。

The images in jpg format are stored in the images directory for training, the labels directory contains annotation files in YOLO format, the annotations directory contains annotation files in VOC format, and the coco directory contains annotation files in COCO format.

数据集下载地址 Download Address 

百度网盘：

链接：https://pan.baidu.com/s/17dRa-CvrX75jiw_Y0RHzbQ?pwd=f3rq

提取码：f3rq

Google Netdisk:

link: https://drive.google.com/file/d/1iwC_ITv2_x1vZFQ7Z8ZnL_kSXMB3DrOa/view?usp=sharing

参考文献 Reference：

[1] Shao, Z., Wu, W., Wang, Z., Du, W., Li, C.: Seaships: A large-scale precisely annotated dataset for ship detection. IEEE transactions on multimedia 20(10), 2593–2604 (2018)

[2] Prasad, D.K., Rajan, D., Rachmawati, L., Rajabally, E., Quek, C.: Video processing from electro-optical sensors for object detection and tracking in a maritime environment: A survey. IEEE Transactions on Intelligent Transportation Systems 18(8), 1993–2016 (2017)