Shore Livecams: A Maritime Dataset for Deep Learning based Object Detection

A maritime dataset for object detection. The dataset is a collection of high definition (HD), full high definition (FHD), ultra high definition (UHD) images captured from live video feeds recorded across various port based areas of Germany. These images contain multiple instances of objects which are primarily classified into three different classes and annotated accordingly.

Prepaired to be used with YOLO PyTorch implementations.

Label formats

Labels are provided in folloing formats:

• YOLO
• PascalVOC .xml (Foldername: train_voc)
• COCO .json

Folder structure

The folder structure looks like this:

images/
    test/
        2ux5cw8z_1080_2022-07-16_11-26_output_0.jpeg
        ...
    train/
        2ux5cw8z_1080_2022-07-17_02-54_output_0.jpeg
        ...
labels/
    test/
        2ux5cw8z_1080_2022-07-16_11-26_output_0.txt
        ...
    test_voc/
        2ux5cw8z_1080_2022-07-16_11-26_output_0.xml
        ...
    train/
        2ux5cw8z_1080_2022-07-17_02-54_output_0.txt
        ...
    train_voc/
        2ux5cw8z_1080_2022-07-17_02-54_output_0.xml
        ...
    shore_livecams.yaml # YOLO v5 PyTorch dataset file for dataloader
    train.json
    test.json
    train+test.json

Dataset Statistics

• 3 classes [human, land, sea]
• 17,6 objects per image average
• Image resolution:
  • 720p (116)
  • 1080p (392)
  • 3840p (17)

Distribution in total, train and test

	images	objects	human	land	sea
total	525	9243	2821	3378	3044
train	394	5643	1624	1976	2043
test	131	3600	1197	1402	1001

Citation

If you find this dataset helpful, please cite this paper (to be published):

D. Ahlers, P. Bhattacharya, P. Nowak, and U. Zölzer, "Shore Livecams: A Maritime Dataset for Deep Learning based Object Detection", in 17th International Conference on Signal Image Technology & Internet based Systems (SITIS), 2023.