Dhaka Traffic Detection with YOLOv5, Streamlit, and FastAPI

This project is a traffic detection system built using YOLOv5, Streamlit, FastAPI, and PyTorch. The goal is to detect traffic objects in images using a trained YOLOv5 model, with options to deploy via Streamlit or FastAPI.

Project Structure

• streamlit/: Streamlit app to upload images and run inference.
• src/: Core Python scripts for data preparation, model training, and inference.
• saved_models/: Folder to store YOLOv5 trained model weights.
• data/: Dataset for training and evaluation.
• uploads/: Uploaded images for inference.
• templates/: HTML templates for the FastAPI frontend.
• static/: Static files (e.g., images) for the FastAPI frontend.

How to Run

Set up the Conda Environment:

conda env create -f environment.yml
conda activate dhaka-traffic-detection

Running the Streamlit App:

To run the Streamlit app locally:

 streamlit run your_app.py --server.runOnSave=false --global.developmentMode=false  

Your app will be available locally at http://localhost:8501.

Running the FastAPI App:

To run the FastAPI app locally:

uvicorn main:app --reload --port 8080

This will start the FastAPI server locally at http://127.0.0.1:8080.

Go to the http://localhost:8000/ - to start the frontend

FastAPI Endpoints

• GET /: Renders the HTML page to upload an image for traffic detection.
• POST /upload/: Uploads an image and returns the processed image with detected objects.

Example to upload an image using curl:

curl -X 'POST' \
  'http://127.0.0.1:8080/upload/' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@your_image.jpg;type=image/jpeg'

Deploy on Streamlit Cloud

Push the project to GitHub and deploy it to Streamlit Cloud using the following steps:

1. Push your project to GitHub.
2. Go to Streamlit Cloud and create a new app.
3. Link it to your GitHub repository and select streamlit_app.py as the main file.
4. Deploy your app to make it accessible online.

Dependencies

• Python 3.9
• Streamlit
• FastAPI
• Uvicorn
• PyTorch
• YOLOv5
• Pillow
• streamlit
• kaggle
• opencv-python
• torch
• python-multipart

Running the Project Locally

1. Activate your Conda environment:

conda activate dhaka-traffic-detection

2. Download the Dataset from Kaggle:

This project uses a dataset hosted on Kaggle. Follow these steps to download the dataset:

• Install the Kaggle API (if not already installed):

pip install kaggle

• Set up the Kaggle API credentials:
  • Download the kaggle.json file from your Kaggle account by navigating to the API section and creating a new API token.
  • Move the kaggle.json file to the appropriate location:

mkdir ~/.kaggle
mv ~/Downloads/kaggle.json ~/.kaggle/
chmod 600 ~/.kaggle/kaggle.json

• Download the dataset: Use the following command to download the YOLOv5 weights dataset into the saved_models/ folder:

kaggle datasets download -d aifahim/dhaka-ai-traffic-challenge-weights-yolov5 -p saved_models --unzip

This will download and unzip the trained YOLOv5 weights into the saved_models/ folder.

Deploying to Streamlit Cloud

1. Push your project to a GitHub repository.
2. Go to Streamlit Cloud and create a new app.
3. Link it to your GitHub repository and select streamlit/app.py as the main file.
4. Deploy your app to make it accessible online!

Additional Notes

• For FastAPI, ensure that you have installed uvicorn, and run the app using the command provided in the "Running the FastAPI App" section.
• This setup will handle everything from model inference to deployment, both using Streamlit and FastAPI.