DeepFish Fish Habitat Dataset

DeepFish is a large-scale benchmark dataset for training and testing several computer vision tasks. The dataset consists of approximately 40,000 images collected underwater from 20 fish habitats in the tropical marine environment of Australia.

The dataset contains classification annotations as well as point-level and segmentation annotations to build a more comprehensive fish analysis benchmark. These annotations enable the model to learn to automatically detect the number of fish, identify their locations, and estimate their sizes.