synthetic-dataset

This is a temporary dataset to organize research on the change-detection project in data for good season 10. It will gather our ideas to create a synthetic change detection dataset and our solutions to detect them. The content of this repo will be shared between the pyro-vision and pyro-dataset repo when we have a stable solution

Install Package

Install synthetic dataset package using:

pip install -e .

Finally you need to add videos of background and smoke in dedicated folders :

• videos\background
• videos\smoke

Create dataset

If you want to generate 10 set please use

python scripts/make_dataset.py --set 10

If you whant all possible combinaisons:

python scripts/make_dataset.py

Train smoke detection network

We provide a notebook to train a smoke detection network using our synthetic dataset

First add pyronear.yaml file to your dataset and then zip it:

cp pyronear.yaml pyro_dataset/pyronear.yaml
zip -r pyro_dataset.zip pyro_dataset

Then store your dataset to your google drive

Finally, use our notebook to train a yolov5 using your dataset

Unitest

You can run unitest using:

coverage run -m unittest discover test/

then check coverage using

coverage html