Money Counter AI

An AI tool that counts coins using computer vision

Video Demo

See YouTube video Build A Computer Vision App from Start to Finish

Setup

Clone this repo

git clone https://github.com/Uniqcoda/money-app

Download raw images from Google Drive and move to the raw_images folder

Google Drive

Enter the directory or open in your favorite code editor

cd money-app

Create a virtual environment. The common way to name virtual environments is env. But you can give it a different name if you need to identify it elsewhere, like on Jupyter notebook or jupyterLab. So, we would use .coin_env instead.

python3 -m venv .coin_env

Add this virtual environment to your .gitignore file, under the Environments section, so you don't push it to GitHub.

Activate the virtual env

source .coin_env/bin/activate # Linux/Mac
.\.coin_env\Scripts\activate # Windows 

Upgrade pip

python3 -m pip install --upgrade pip

If working on Jupyter notebook, and you need the virtual env, run this code in your terminal, then go to jupyter nb and select the virtual env from the list of kernels.

pip install ipykernel
python3 -m ipykernel install --user --name=coin_env

On Jupyter notebook

On JupyterLab

Image Preprocessing

Video reference

1. Install Pillow (on the virtual env)

python3 -m pip install --upgrade Pillow

2. Run script to resize and crop images

Add collected images to the raw_images folder

If running from vscode, run

python3  preprocess.py

If on Jupyter notebook, run the preprocess_images.ipynb file

Your cleaned images should be in the /images folder.

Image Upload & Labelling

=> This step would be done on the GCP VertexAI platform. Follow the image upload and labelling video for further instructions.
=> Also check out Vertex AI documentation on object detection. This docs is also applicable to model training, evaluation, deployment and testing.

Model Training, Evaluation & Deployment

This step is also done on GCP VertexAI platform. Follow the Model Training video for further instructions.

Web App Setup

We would be using the endpoint from the deployed model in the previous step to create predictions for the web app.

1. Install dependencies

pip install Flask google-cloud-aiplatform python-dotenv

2. Create a new file named .env, under the root directory (main folder). Add the API keys and other environment variables to the .env file.

PROJECT_NUMBER=0000000000
LOCATION=europe-west4
ENDPOINT_ID=0000000000
API_ENDPOINT=europe-west4-aiplatform.googleapis.com

3. Google cloud service account authentication Ref: https://developers.google.com/identity/protocols/oauth2/service-account#python
   Follow the setup instructions to (i) create a service account and (ii) create a service account key.
   The new service account key will be downloaded to your computer. Move it to the directory/folder where you need it. Then add the file path to the script below and run

export GOOGLE_APPLICATION_CREDENTIALS="path/to/downloaded_key.json"

4. Start the Flask app

flask run

Sample model prediction result for single object detection

python3 test-prediction.py

Result

{
    "bboxes": [[0.316620409, 0.678313732, 0.435367882, 0.789723396]],
    "confidences": [0.994227529], 
    "displayNames": ["2p"], 
    "ids": ["2112341057353023488"]
}

Docs: https://cloud.google.com/vertex-ai/docs/image-data/object-detection/interpret-results

5. Final app