SEA-AI/SEANet

SEADs: Maritime Multimodal Panoptic Segmentation Dataset 🌊🚢

Dataset Summary

This dataset contains multimodal maritime imagery for panoptic segmentation and object detection tasks.
It includes aligned RGB and thermal images, with merged panoptic annotations in COCO format.

The dataset is particularly designed for tasks such as:

• Vessel and obstacle detection
• Water-sky-land segmentation
• Multimodal feature fusion (RGB + Thermal)
• Horizon detection

⚠️ Status: Work in Progress
This dataset is still under construction.

• We are continuously adding more images
• Polishing the annotation format (COCO Panoptic standard)
• Improving mask quality and consistency

Please expect changes in structure and annotations over time, we wil let you know once the dataset is fully finished!

Dataset Structure

dataset/
│── images/
│ ├── rgb/ # RGB images
│   ├── train/
│   ├── val/
│ ├── thermal/ # Thermal images (aligned with RGB)
│   ├── train/
│   ├── val/
│── annotations/
│ ├── panoptic_train/
│ ├── panoptic_val/
│ ├── panoptic_train.json
│ └── panoptic_val.json

• Each sample has both RGB and Thermal inputs
• A single merged panoptic annotation (union of RGB + thermal labels)

Usage

Citation

License

CC BY-NC 4.0

Contact

For questions or contributions, please reach out to: [email protected]