TUD-GV Dataset for Floating Litter Detection

This dataset contains the data used for the publication:

Jia T, Vallendar AJ, de Vries R, Kapelan Z and Taormina R (2023) Advancing deep learning-based detection of floating litter using a novel open dataset. Front. Water 5:1298465. https://doi.org/10.3389/frwa.2023.1298465

The large-scale "TU Delft - Green Village" (TUD-GV) dataset is for detecting floating litter with computer vision. We created this dataset from experiments conducted during 10 days in February and April 2021 in a small drainage canal at The Green Village — a field lab facility in the TU Delft Campus, the Netherlands. We captured data using two action cameras (GoPro HERO4 and GoPro MAX 360) and a phone (Huawei P30 Pro) mounted on four different locations on a bridge. All devices recorded videos with a resolution of 1080p, a linear field of view, and a FPS (frame per second) of 24 (for the action cameras) or 30 (for the phone).

This dataset consists of 9,473 RGB images. We manually labeled the images in the TUD-GV dataset into four classes: no litter (0 items), little litter (1-2 items), moderate litter (3-5 items), and lots of litter (6-10 items) according to the number of litter items in images.

This dataset is stored in the ZIP file contain 77 directories and the TUD-GV.xls file. Each of these directories contain four class label directories, and each label directory contains JPG images. The TUD-GV.xls file contains the detailed information of images in 77 directories, including collecting date, collecting time, device, device location (in a bridge), device degree, device height, weather conditions, litter source, the number of images per class, and the number of images in total.

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

@article{jia2023advancing,
title={Advancing deep learning-based detection of floating litter using a novel open dataset},
author={Jia, Tianlong and Vallendar, Andre Jehan and de Vries, Rinze and Kapelan, Zoran and Taormina, Riccardo},
journal={Frontiers in Water},
volume={5},
pages={1298465},
year={2023},
publisher={Frontiers Media SA}
}