This dataset is proposed by the ICML 2024 paper "Diving into Underwater: Segment Anything Model Guided Underwater Salient Instance Segmentation and A Large-scale Dataset", specific parameters about the dataset can be viewed in the paper
In this paper, we construct the first large-scale USIS10K dataset for the underwater salient instance segmentation task, which contains 10,632 images and pixel-level annotations of 7 categories. As far as we know, this is the largest salient instance… See the full description on the dataset page:
https://huggingface.co/datasets/LiamLian0727/USIS10K.
