wangbo66/STAFDD-dataset

STAFDD Dataset

Overview

The STAFDD Dataset is a fish disease detection dataset designed for training and evaluating deep learning models under aquaculture scenarios.
It supports object detection, spatio-temporal analysis, and video-based disease assessment, and is used in the paper:

STAFDD: A Spatio-Temporal Automatic Fish Disease Detection Method

The dataset is organized in YOLO format and includes:

• Annotated images for YOLO-based training
• Pretrained .pt model weights
• Raw test videos for inference and evaluation

---

Dataset Structure

The dataset is organized as follows:

STAFDD-dataset/ ├── images/ │ ├── train/ │ ├── val/ │ └── test/ ├── labels/ │ ├── train/ │ ├── val/ │ └── test/ ├── videos/ │ └── test_videos/ ├── models/ │ └── ReID.pt ├── data.yaml └── README.md

• images/: RGB images extracted from aquaculture videos
• labels/: YOLO-format annotations (.txt)
• videos/: Raw test videos used for model evaluation
• models/: Pretrained model weights (.pt)
• data.yaml: YOLO training configuration file

---

Annotation Format

Annotations follow the YOLO object detection format:

• Coordinates are normalized to [0, 1]
• One annotation file per image
• Bounding boxes correspond to visible disease-related regions on fish bodies

---

Classes

The dataset focuses on fish disease-related visual symptoms.
Class definitions are consistent with those described in the associated paper.

⚠️ Note: Class semantics should be interpreted together with the experimental section of the paper.

---

Data Splits

The dataset is divided into three subsets:

• Training set
• Validation set
• Test set

To reduce data leakage, frame sampling and dataset splitting are performed with temporal consistency considerations, avoiding random shuffling of adjacent video frames.

---

Intended Use

This dataset is intended for:

• Fish disease detection
• Small object detection in aquaculture environments
• Video-based disease analysis
• Research on spatio-temporal fish health monitoring

It is suitable for training YOLO-based detectors and evaluating models on real-world aquaculture videos.

---

License

This dataset is released under the CC BY 4.0 license.

---

Citation

If you use this dataset, please cite the following paper:

@article{wang2024stafdd,
  title={STAFDD: A Spatio-Temporal Automatic Fish Disease Detection Method},
  author={Wang, Bo and others},
  journal={},
  year={2024}
}


Contact

For questions or collaborations, please contact:

Bo Wang
Email: 3020201781@jsnu.edu