Image detection AI with a neural networks to detect some sodas.
- Install Python 3.11 via Windows Store
- Clone the project from GitHub and open it via PyCharm
- Set up the virtual environment and do not forget to activate it before running the project
- Install the required packages via
pip install -r requirements.txt- On windows, type this command
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124and also do not forget to install Cuda. - https://pytorch.org/get-started/locally/#start-locally for more information on which PyTorch version to install
- On windows, type this command
- Copy
.env.exampleto.envand setup your environment variables - Run the project via
python main.py
You can find the API Token of Picsellia in Personal Settings > Tokens
- Install pre-commit via
pip install pre-commit(if not already installed via the requirements.txt) - Run the command
pre-commit installin the project directory - Now, every time you commit, the pre-commit hooks will be executed
You can manually run the pre-commit hooks via
pre-commit run --all-files
- Python 3.11
- PyCharm
- Cuda for Windows
Run either the training or the inference command ! :)
If you are lost, run the command
python main.py infer -horpython main.py train -hto get the help message
Here is the usage version
usage: main.py train [-h] dataset_version_id project_id
positional arguments:
dataset_version_id Version ID of the Picsellia dataset to use for training
project_id Picsellia Project ID to use for training
options:
-h, --help show this help message and exit
Here is the usage version to run the inference with an image, a video or open the webcam to detect the items
usage: main.py infer [-h] (--image IMAGE | --video VIDEO | --webcam) [--output OUTPUT] [--conf CONF] [--iou IOU] model_version_id
positional arguments:
model_version_id Version ID of the model to use for inference
options:
-h, --help show this help message and exit
--image IMAGE Path to the image for inference
--video VIDEO Path to the video for inference
--webcam Use webcam for inference
--output OUTPUT Path to save inference results (ex: output directory for annotated images/videos)
--conf CONF Confidence threshold for detections (default: 0.5)
--iou IOU IoU threshold for NMS (default: 0.45)
Examples :
python main.py infer --video 'video.mp4' --output output --conf 0.9 --iou <0.5 01943c53-956c-775b-afd1-38e9f87ed22e>
python main.py infer --webcam --conf 0.9 <01943c53-956c-775b-afd1-38e9f87ed22e>
Note
IoU (Intersection over Union) measures how accurately a predicted bounding box overlaps the ground truth, while confidence reflects the model's certainty in its prediction.