Oostpoort Dataset for Floating Litter Detection

This dataset contains the data used for the publication:

Jia, T., de Vries, R., Kapelan, Z., van Emmerik, T. H., & Taormina, R. (2024). Detecting floating litter in freshwater bodies with semi-supervised deep learning. Water Research, 266, 122405.

The Oostpoort dataset is for detecting floating litter with computer vision. We generated this dataset from experiments conducted during 26 days from February to March 2022, in a canal at Oostpoort, Delft, the Netherlands. We collected data employing action cameras (GoPro MAX 360 and GoCam3) mounted outside the windows of a tower at Oostpoort with a viewing angle of 0 degree. We recorded video sequences with a time-lapse recording (1 image/30 sec) and a FPS (frame per second) of 17.98. We generated the Oostpoort dataset by saving images from these videos. The resolution of images are 3840*2160 and 1920*1440. This dataset consists of 562 RGB images. We manually labeled the litter items in these images with bounding boxes.

The 562 images are stored in the images.zip file, the annotations are stored in the labels_txt.zip file, and the class of the annotation (i.e., litter) is stored in the classes.txt file. The Oostpoort Dataset.xlsx file contains the detailed information of images, including collecting date, collecting time, device, device location (in a bridge), device degree, device height, weather conditions, the number of images, and the number of annotated litter items.

If you use this dataset for a publication, please cite the paper. Here is a BibTeX entry:

@article{jia2024detecting,
title={Detecting floating litter in freshwater bodies with semi-supervised deep learning},
author={Jia, Tianlong and de Vries, Rinze and Kapelan, Zoran and van Emmerik, Tim HM and Taormina, Riccardo},
journal={Water Research},
volume={266},
pages={122405},
year={2024},
publisher={Elsevier}
}