A multimodal dataset for perception in real-world maritime navigation scenarios

MarScenes3D is a multimodal maritime perception dataset collected from a moving vessel in real-world navigation scenarios, including coastal waterways, ports, anchorages, and open navigation routes. It provides synchronized surround-view images, LiDAR point clouds, ego-vessel motion data, and sensor calibration parameters, together with 2D and 3D object annotations, tracking identities, and structured language descriptions. The dataset can support research on maritime object detection, multi-object tracking, and multimodal scene understanding.