Underwater datasets for MADNet

The underwater datasets used for MADNet are Brackish, UOv2, and RUOD.File: UOv2.zipDescription: It contains five target categories: echinus, holothurian, scallop, starfish, and waterweeds. The dataset is divided into a training set of 5,320 samples, a validation set of 1,520 samples, and a test set of 760 samples, following a 7:2:1 ratio.File: Brackish.zipDescription: The dataset, which contains 14,674 images of crabs, fish, jellyfish, shrimp, small fish, and starfish, was randomly divided into a training set, a validation set, and a test set in an 8:1:1 ratio, comprising 11,739, 1,467, and 1,468 images, respectively.File: RUOD.zipDescription: It contains 14,000 images with 10 target categories, namely holothurian, echinus, scallop, starfish, fish, corals, diver, cuttlefish, turtle, and jellyfish. It is divided into 9,800 images with 51,934 annotations for the training set and 4,200 images with 22,969 annotations for the validation set, respectively. A total of 300 images were provided in the test set, comprising 100 images each for blur, color, and light.Data was derived from the following sources:Brackish: https://universe.roboflow.com/brad-dwyer/brackish-underwaterUOv2: https://universe.roboflow.com/roboflow-100/underwater-objects-5v7p8RUOD: https://drive.google.com/file/d/1hxtbdgfVveUm_DJk5QXkNLokSCTa_E5o/view?usp=drive_link