eStonefish-scenes: A synthetically generated dataset for underwater event-based optical flow prediction

The eStonefish-Scenes dataset is a synthetic event-based dataset meticulously designed to simulate diverse underwater environments under varying conditions using the Stonefish simulator. The dataset is specifically tailored for autonomous underwater vehicles (AUVs), featuring event streams, grayscale images, and corresponding ground truth optical flow.

In addition to the dataset, the accompanying repository provides a user-friendly pipeline for generating custom underwater event-based datasets, which includes optical flow displacements and grayscale images. The generated data leverages the optimized eWiz framework, ensuring efficient storage, access, and processing.

The data generation pipeline can be utilized by cloning the estonefish-scenes repository. We also include stonefish-scenegen, which automatically generates realistic underwater scenes for use with Stonefish. The goal behind stonefish-scenegen is to simplify the creation of Stonefish environments, by randomly placing clusters of coral models on a seabed to create diverse and natural-looking environments.

The eStonefish-scenes dataset follows the eWiz format, which stores data in HDF5 files. Specifically, events.hdf5 holds the event streams, gray.hdf5 contains grayscale image sequences associated with each event stream, and flow.hdf5 stores the optical flow information. Metadata and configuration details, including sensor specifications and recording parameters, are saved in props.json. This structure facilitates straightforward access to each data type, improving data handling and compatibility with various processing pipelines. The directory structure is summarized below:

dataset_root/├── events.hdf5├── gray.hdf5├── flow.hdf5└── props.json

Whether you're a researcher or developer, this resource is an ideal starting point for advancing event-based vision systems in real-world underwater applications.