The Maritime Synthetic (MarSyn) dataset consists of 25 different photo-realistic video sequences, containing each one 1000 frames, obtained using Blender. Its 25 thousand images and their corresponding 34 thousand annotations aim to simulate multiple maritime scenarios and conditions such as sunny, cloudy and sunset environments, near coast images and reflections on the water. Different types of vessels are deployed in the simulations, which increases the diversity of the dataset. The vessels vary in type (cargo ships, military ships, fishing boats, speed boats, rescue rafts and others), length (from 3 meters all the way up to 125 meters), shape and color. The simulated camera is placed and moved accordingly to an UAV and it captures imagery at heights between 150 and 1000 meters. All of the captured images are RGBA and are available in two different resolutions, 1280×720 pixels and 550×550 pixels.
This method of maritime images generation allows the creation of rich and diverse maritime datasets, with different scenarios and configurations, removing the costs of real-world data acquisition and making it possible to generate specific situations that can be hard to capture in real-world scenarios: in this way, it is possible to train a network to segment different types of vessels without having to capture them in the real world. Also, by producing the foreground masks automatically, no manual labeling is required, reducing time and human errors.
For further information about the MarSyn Dataset, please contact .