We train on a combination of several datasets. If you execute the training script with the --download
parameter, all
necessary datasets should be downloaded automatically
we re-downloaded the Scenes from Google Earth Engine in the scenes
subfolder
please cite the original MARIDA paper: https://github.com/marine-debris/marine-debris.github.io
Kikaki K, Kakogeorgiou I, Mikeli P, Raitsos DE, Karantzalos K (2022) MARIDA: A benchmark for Marine Debris detection from Sentinel-2 remote sensing data. PLoS ONE 17(1): e0262247. https://doi.org/10.1371/journal.pone.0262247
download the original MARIDA data from here: https://doi.org/10.5281/zenodo.5151941
We added Sen2Cor corrected Sentinel-2 scenes in this modified dataset (21GB compressed; 24GB unzipped).
dataset available here https://github.com/ESA-PhiLab/floatingobjects this link: https://drive.google.com/drive/folders/1QGjzRTVRQbf4YbzfUWMeIdJvYkzuipGJ?usp=sharing
please cite the authors for this dataset
Mifdal, J., Longépé, N., and Rußwurm, M.: TOWARDS DETECTING FLOATING OBJECTS ON A GLOBAL SCALE WITH LEARNED SPATIAL FEATURES USING SENTINEL 2, ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-3-2021, 285–293, https://doi.org/10.5194/isprs-annals-V-3-2021-285-2021, 2021.
Carmo, R., Mifdal, J., and Rußwurm, M.: Detecting Macro Floating Objects on Coastal Water Bodies using Sentinel-2 Data, OCEANS 2021 San Diego – Porto, 2021.
we mirror the dataset here (18GB compressed; 37GB unzipped)
In this work, we annotated additional scenes from the FloatingObjects dataset by accurate points. The dataset is available here: https://marinedebrisdetector.s3.eu-central-1.amazonaws.com/data/refinedfloatingobjects.zip (1.3GB compressed; 12GB unzipped)
The targets were deployed during the Plastic Litter Projects 2021 and 2022
Papageorgiou, D., Topouzelis, K., Suaria, G., Aliani, S., Corradi, P., 2022. Sentinel-2 detection of floating marine litter targets with partial spectral unmixing and spectral comparison with other floating materials (plastic litter project 2021) Under review.
We acquired and annotated the Sentinel-2 scenes here: https://marinedebrisdetector.s3.eu-central-1.amazonaws.com/data/PLP.zip (38MB compressed; 41MB unzipped)
We use examples of ships as negatives to teach the model to not confuse ships with marine debris. To do so, we used data from this repository: https://github.com/alina2204/contrastive_SSL_ship_detection and paper https://www.mdpi.com/2072-4292/13/21/4255/htm
please cite the original authors
Ciocarlan, A., & Stoian, A. (2021). Ship Detection in Sentinel 2 Multi-Spectral Images with Self-Supervised Learning. Remote Sensing, 13(21), 4255.
we only used the Sentinel-2 scenes here (685MB compressed; 2.6GB unzipped)
We additionally annotated the durban scene to measure confusions with other objects. The data is available here durban scene available here: https://marinedebrisdetector.s3.eu-central-1.amazonaws.com/data/durban.zip