Aquatic plastic litter dataset developed for APLASTIC-Q publication

The two datasets were created for the publication 'Machine learning for aquatic plastic litter detection, classification and quantification (APLASTIC-Q)' by Wolf et. al 2020. It consists of a Plastic Litter Detector (PLD) dataset which contains classes with litter present (two classes) and classes with litter-free areas (four classes). The second Plastic Litter Quantifier (PLQ) dataset contains classes of litter types (14 classes) and classes of litter-free areas (4 classes). Each data set is split into a train (80%) and a test dataset (20%).

The datasets are stored in the ZIP file contain two directories (PLD and PLQ datasets) which contain two directories each (train and test datasets). Each of these subdirectories contain the class label directories (subdirectories of PLD dataset has 6 label directories, PLQ dataset has 18 label directories). Each of the label directories contain PNG image tiles.

PLD-dataset consists of 6,892 PNG true color RGB image tiles with the shape of 100x100x3. The class labels along with number of total samples (train and test dataset) are: litter classes Litter – high (1,905) and Litter – low (1,146), litter-free classes Water (1,042), Sand (602), Vegetation (1,840) and Other (357).

PLQ-dataset consists of 6,026 PNG true color RGB image tiles with the shape of 50x50x3. The class labels along with number of total samples are: litter type classes Cans (42), Carton (26), Plastic bag – large (684), Plastic bag – small (493), Plastic bottles (878), Plastic bowls (96), Plastic canister (16), Plastic cups (26), Plastic other (36), Polystyrene packaging (693), Shoes (8), String and cord (12), Styrofoam (614) and Textiles (16), litter-free classes Water (935), Sand (467), Vegetation (752) and Other (232).