hefy/pyro-sdis

---
license: apache-2.0
dataset_info:
features:
- name: image
dtype: image
- name: annotations
dtype: string
- name: image_name
dtype: string
- name: partner
dtype: string
- name: camera
dtype: string
- name: date
dtype: string
splits:
- name: train
num_bytes: 2940743706.011
num_examples: 29537
- name: val
num_bytes: 391545545.068
num_examples: 4099
download_size: 3284043758
dataset_size: 3332289251.079
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: val
path: data/val-*
tags:
- wildfire
- smoke
- yolo
- pyronear
- ultralytics
size_categories:
- 10K<n<100K
---

# Pyro-SDIS Dataset


---

## About the Dataset

Pyro-SDIS is a dataset designed for wildfire smoke detection using AI models. It is developed in collaboration with the Fire and Rescue Services (SDIS) in France and the dedicated volunteers of the Pyronear association.

The images in this dataset come from Pyronear cameras installed with the support of our SDIS partners. These images have been carefully annotated by Pyronear volunteers, whose tireless efforts we deeply appreciate.

We extend our heartfelt thanks to all Pyronear volunteers and our SDIS partners for their trust and support:

- **Force 06**
- **SDIS 07**
- **SDIS 12**
- **SDIS 77**

Additionally, we express our gratitude to the DINUM for their financial and strategic support through the AIC, Etalab, and the Legal Service. Special thanks also go to the Mission Stratégie Prospective (MSP) for their guidance and collaboration.

The Pyro-SDIS Subset contains **33,636 images**, including:

- **28,103 images with smoke**
- **31,975 smoke instances**

This dataset is formatted to be compatible with the Ultralytics YOLO framework, enabling efficient training of object detection models.

---

Stay tuned for the full release in **January 2025**, which will include additional images and refined annotations. Thank you for your interest and support in advancing wildfire detection technologies!

## Dataset Overview

### Contents
The Pyro-SDIS Subset contains images and annotations for wildfire smoke detection. The dataset is structured with the following metadata for each image:

- **Image Path**: File path to the image.
- **Annotations**: YOLO-format bounding box annotations for smoke detection:
- `class_id`: Class label (e.g., smoke).
- `x_center`, `y_center`: Normalized center coordinates of the bounding box.
- `width`, `height`: Normalized width and height of the bounding box.
- **Metadata**:
- `partner`: Partner organization responsible for the camera (e.g., SDIS 07, Force 06).
- `camera`: Camera identifier.
- `date`: Date of image capture (formatted as `YYYY-MM-DDTHH-MM-SS`).
- `image_name`: Original file name of the image.
- **Split**: Indicates whether the image belongs to the training or validation set (`train` or `val`).

### Example Record
Each record in the dataset contains the following structure:
```json
{
"image": "./images/train/partner_camera_date.jpg",
"annotations": "0 0.5 0.5 0.1 0.2",
"split": "train",
"image_name": "partner_camera_date.jpg",
"partner": "partner",
"camera": "camera",
"date": "YYYY-MM-DDTHH-MM-SS"
}
```

---

Let me know if you’d like further refinements or if you want me to include specific numbers/statistics for the dataset.

### Splits
The dataset is divided into:
- **Training split**: Used for training the model.
- **Validation split**: Used to evaluate model performance.

## Exporting the Dataset for Ultralytics Training

To train a YOLO model using the Ultralytics framework, the dataset must be structured as follows:
- **Images**: Stored in `images/train/` and `images/val/` directories.
- **Annotations**: Stored in YOLO-compatible format in `labels/train/` and `labels/val/` directories.

### Steps to Export the Dataset

1. **Install Required Libraries**:
```bash
pip install datasets ultralytics
```

2. **Define Paths**:
Set up the directory structure for the Ultralytics dataset:
```python
import os
from datasets import load_dataset

# Define paths
REPO_ID = "pyronear/pyro-sdis"
OUTPUT_DIR = "./pyro-sdis"
IMAGE_DIR = os.path.join(OUTPUT_DIR, "images")
LABEL_DIR = IMAGE_DIR.replace("images", "labels")

# Create the directory structure
for split in ["train", "val"]:
os.makedirs(os.path.join(IMAGE_DIR, split), exist_ok=True)
os.makedirs(os.path.join(LABEL_DIR, split), exist_ok=True)

# Load the dataset from the Hugging Face Hub
dataset = load_dataset(REPO_ID)
```

3. **Export Dataset**:
Use the following function to save the dataset in Ultralytics format:
```python
def save_ultralytics_format(dataset_split, split):
"""
Save a dataset split into the Ultralytics format.
Args:
dataset_split: The dataset split (e.g., dataset["train"])
split: "train" or "val"
"""
for example in dataset_split:
# Save the image to the appropriate folder
image = example["image"] # PIL.Image.Image
image_name = example["image_name"] # Original file name
output_image_path = os.path.join(IMAGE_DIR, split, image_name)

# Save the image object to disk
image.save(output_image_path)

# Save label
annotations = example["annotations"]
label_name = image_name.replace(".jpg", ".txt").replace(".png", ".txt")
output_label_path = os.path.join(LABEL_DIR, split, label_name)

with open(output_label_path, "w") as label_file:
label_file.write(annotations)

# Save train and validation splits
save_ultralytics_format(dataset["train"], "train")
save_ultralytics_format(dataset["val"], "val")

print("Dataset exported to Ultralytics format.")
```

4. **Directory Structure**:
After running the script, the dataset will have the following structure:
```
pyro-sdis/
├── images/
│ ├── train/
│ ├── val/
├── labels/
│ ├── train/
│ ├── val/
```

---

### Training with Ultralytics YOLO

1. **Download the `data.yaml` File**:
Use the following code to download the configuration file:
```python
from huggingface_hub import hf_hub_download

# Correctly set repo_id and repo_type
repo_id = "pyronear/pyro-sdis"
filename = "data.yaml"

# Download data.yaml to the current directory
yaml_path = hf_hub_download(repo_id=repo_id, filename=filename, repo_type="dataset", local_dir=".")
print(f"data.yaml downloaded to: {yaml_path}")
```

2. **Train the Model**:
Install the Ultralytics YOLO framework and train the model:
```bash
pip install ultralytics
yolo task=detect mode=train data=data.yaml model=yolov8n.pt epochs=50 imgsz=640 single_cls=True
```

## License

The dataset is released under the [Apache-2.0 License](https://www.apache.org/licenses/LICENSE-2.0).

## Citation

If you use this dataset, please cite:
```
@dataset{pyro-sdis,
author = {Pyronear Team},
title = {Pyro-SDIS Dataset},
year = {2024},
publisher = {Hugging Face},
url = {https://huggingface.co/pyronear/pyro-sdis}
}
```