TUD-HCMC Dataset for Floating Litter Detection and Flux Quanitification

Dataset This dataset contains the data used for the publication: Jia, T., Taormina, R., de Vries, R., Kapelan, Z., van Emmerik, T. H., Vriend, P., & Okkerman, I. (2025). A semi-supervised learning-based framework for quantifying litter fluxes in river systems. Water Research, 124833. The TU Delft - Ho Chi Minh City (TUD-HCMC) dataset is for detecting floating litter and quantifying litter fluxes in rivers with wide cross-sections using computer vision techniques. The Data information.xlsx file shows the detailed information of the collected images. The images and annotations are stored in two folders (i.e., Thu Thiem 0909 and Binh Loi 1209) We conducted measurements at the Binh Loi and Thu Thiem bridges across the Saigon River in Ho Chi Minh City, Vietnam, over two days during the wet season in September 2023. We divided each bridge into five transects, and monitored floating litter at the center of each transect. The length of these transects was carefully selected to ensure that the bridge piers were not visible within the camera's field of view during sampling. All measurements were performed in the southernmost side of bridges during the ebb tide. On each measurement day, we conducted 4 or 6 rounds of measurements. During each round, we captured images (6016×4000 pixels) sequentially from the sampling point 1 to 5, using a handheld camera (Pentax K-series) over a period of 130 seconds. The camera was oriented nearly vertically with respect to the water surface, with a time-lapse recording (1 image/10 seconds). The observation area width for each sampling point is 7 m, and the ground sampling distance (GSD) of each image is 0.12 cm/pixel. Due to instability in the operation of the handheld camera, we only selected the images without heavy blur for measuring litter fluxes. Finally, we built the TUD-HCMC dataset, including 199 images and 309 images collected from the Thu Thiem and Binh Loi bridge, respectively. We annotated litter items in images using bounding boxes to indicate their locations, resulting in 64 and 114 annotations in these two locations, respectively. Since some items appear in multiple consecutive images, the number of annotated litter items exceeds the actual number of litter items in the rivers (see Data information.xlsx file). Cite this dataset If you use this dataset for a publication, please cite the paper. Here is a BibTeX entry: @article{jia2025semi, title={A semi-supervised learning-based framework for quantifying litter fluxes in river systems}, author={Jia, Tianlong and Taormina, Riccardo and de Vries, Rinze and Kapelan, Zoran and van Emmerik, Tim HM and Vriend, Paul and Okkerman, Imke}, journal={Water Research}, pages={124833}, year={2025}, publisher={Elsevier} }