FloW-Img

The FloW-Img sub-dataset contains 2000 images with more than 5000 thousand labeled floating wastes. The number of objects in one frame varies from 1 to 17. Among all the labeled floating wastes, the small objects (which occupy less than 32 × 32 pixels) account for the largest proportion in the dataset.

Besides, we also provide 200 unannotated video sequences of floating waste for further research on floating waste detection.