Efficient methods for underwater coral detection and high-quality underwater coral detection dataset for machine learning and computer vision research.
- [2025/2/9]: We have added 5 coral videos to enrich the dataset and provide additional resources for your research. 🌊
- [2024/12/22]: Added detailed instructions for integrating and using SCoralDet source code within the Ultralytics YOLO framework. 🎉
- [2024/12/18]: We have uploaded the core improvement source code of SCoralDet! 😊
- [2024/12/16]: We are maintaining the SCoralDet Dataset on Papers with Code.
- [2024/12/13]: Our paper SCoralDet: Efficient real-time underwater soft coral detection with YOLO has recently been published! [Paper]
- [2024/12/01]: We’ve added specific instructions on how to use the dataset for your research!
- [2024/11/20]: We've opened up our dataset! [SCoralDet-Dataset]
The SCoralDet Dataset is a high-quality underwater coral dataset, containing 646 annotated images collected from the Coral Germplasm Conservation and Breeding Center at Hainan Tropical Ocean University in Sanya, China. This dataset is specifically designed for coral detection and recognition tasks, making it ideal for researchers and practitioners working on underwater image analysis, marine biology, and computer vision applications.
The dataset includes images from the following six coral species:
- Euphylliaancora
- Favosites
- Platygyra
- Sarcophyton
- Sinularia
- Wavinghand
- Total images: 646
- Format: PNG/JPEG
- Annotations: Bounding boxes in COCO format
- Resolution: Variable (high-resolution images)
You can download the dataset from Google Drive.
We also provide 5 coral videos as part of the dataset. You can access and download them via the following Google Drive.
Here are a few samples from the dataset:
| Euphylliaancora | Platygyra | Sarcophyton |
|---|---|---|
 |
 |
 |
| Wavinghand | Favosites | Sinularia |
|---|---|---|
 |
 |
 |
Building upon the SCoralDet dataset, we developed MambaCoral-DiffDet as a refined and advanced framework for soft coral detection under complex underwater conditions. This work utilizes diffusion models to address data imbalance and improves detection accuracy using novel attention mechanisms and architecture modifications.
For more details, please refer to the MambaCoral-DiffDet, where you can see how SCoralDet has been leveraged and extended in this project.
This dataset can be used for:
- Coral species classification
- Object detection tasks (e.g., YOLO, Faster R-CNN)
- Transfer learning on underwater datasets
To get started with the SCoralDet Dataset, you can download it directly from Google Drive.
Once you have the dataset, unzip the downloaded files. The dataset is organized into the following structure:
/your_dataset_path
├── annotations
└── image
├── Euphflfiaancora
├── Favosites
├── Platygyra
├── Sarcophyton
├── Sinularia
└── WavingHandannotations/: Contains annotation files in COCO format for each image in theimage/directory.image/: Contains images for each of the six coral species, organized in separate subdirectories.
You can use the Dataset_process.ipynb file to process the dataset and get the images in PNG/JPEG format with COCO annotations. You can split the dataset for training and validation.
If you'd like to use the dataset in YOLO format, use the coco2yolo.py script:
python coco2yolo.py --json_path your_coral_instances.json --save_path your_yolo_label_path|-- SCoralDet/
|-- README.md # Documentation for the project
|-- lib/ # Core library containing configurations, modules, and utilities
|-- cfg/
|-- models/
|-- v10/ # YOLOv10 model configurations
|-- yolov10n.yaml # Main YOLOv10n model structure
|-- yolov10n_backbone.yaml # Main YOLOv10n MPFB-backbone configuration
|-- yolov10n_lightneck.yaml # YOLOv10n lightweight neck configuration
|-- nn/
|-- extra_modules/ # Additional model modules
|-- block.py # Includes custom modules: MPFB & LightNeck
|-- utils/
|-- tal.py # Implements APT
Follow the Ultralytics YOLO Framework v8.3.51 guide to set up the environment and dependencies. Install PyTorch:
pip install torch==2.2.0 torchvision==0.17.2 torchaudio==2.2.0 --index-url https://download.pytorch.org/whl/cu121After setting up Ultralytics YOLO, the next step is to add SCoralDet's custom code to the framework. Below are the specific modules and their purposes:
1. Add the `extra_modules/` folder:
└── Copy the `block.py` file into the framework's directory for module extensions.
- Purpose: Implements MPFB (Multi-Path Fusion Block) and LightNeck modules.
2. Add the `utils/tal.py` file:
└── Copy this utility script into the framework.
- Purpose: Provides the APT algorithm.
3. Copy the following configuration files from lib/cfg/models/v10/ into the YOLO framework:
- `yolov10n_backbone.yaml`: Configures the MPFB backbone module.
- `yolov10n_lightneck.yaml`: Defines the lightweight neck structure. Once SCoralDet is integrated, you can begin training the model by executing the following command:
python train.pyAfter training, validate the model’s performance using the following command:
python val.pyIf you find our work useful, please consider citing us or giving us a star ⭐.
@article{lu2024scoraldet,
author={Lu, Zhaoxuan and Liao, Lyuchao and Xie, Xingang and Yuan, Hui},
title={SCoralDet: Efficient real-time underwater soft coral detection with YOLO},
journal={Ecological Informatics},
year={2024},
artnum={102937},
issn={1574-9541},
doi={10.1016/j.ecoinf.2024.102937},
}
@article{lu2024underwater,
title={Underwater Soft Coral Detection: SCoralNet for Accurate and Efficient Annotation.},
author={Lu, Zhaoxuan and Xie, Xingang and Zhu, Xiaolong},
journal={IAENG International Journal of Computer Science},
volume={51},
number={9},
year={2024}
}We welcome contributions to improve this dataset and code! Feel free to:
- Report issues or bugs.
- Submit pull requests for any enhancements.
- Email us directly at [luzhaoxuan@smail.fjut.edu.cn].