J-Litter: An Image Dataset for Deep-Sea Plastic and Macrolitter Detection

This is a dataset designed for training AI to detect plastic and macrolitter in images captured during deep-sea observations. The target categories are "plastic-film," "plastic-bottle," "beverage-can," "other-metal," and "others." The dataset consists of approximately 10,000 image files. These images were extracted from deep-sea videos available on JAMSTEC's "Deep-sea Debris Database" and "Deep-sea Video & Image Archives (J-EDI)," as well as from footage taken during deep-sea surveys by the "Shinkai 6500" and "Hyper-Dolphin" between 2022 and 2025. The dataset includes a JSON file containing bounding boxes and label information created in the COCO (Common Objects in Context) format, a widely used training data format in the image AI field. The entire dataset is available for download as a zip file.