vdrvar / comp_vision_for_soda_bottles

Soda bottle recognition using transfer learning. Achieves 94% accuracy with a fine-tuned Xception model. Features a FastAPI interface for image uploads and predictions. Uses Python and TensorFlow.

Transfer Learning for Soda Bottle Recognition

This project demonstrates the power of transfer learning in computer vision by applying it to identifying soda bottles. We fine-tune a pre-trained Xception model to achieve high accuracy in recognizing various soda bottle brands.

Project Overview:

• Deep Learning in Computer Vision: Uses advanced techniques to accurately recognize soda bottle images.

• Transfer Learning Approach: Applies the Xception model, enhanced through transfer learning, to identify soda bottles.

• Kaggle Dataset: Trains and evaluates the model using the Cola Bottle Identification Dataset from Kaggle.

Key Components:

• /app: Web interface for image upload and prediction.
• /code/conv_net_comparison.ipynb: Documentation of neural network comparisons.
• /data/soda_bottles: Dataset for model training and validation.
• /other_files: Supplementary resources, including weights and test photos.

Quick Start:

1. Clone the repository:

git clone https://github.com/vdrvar/comp_vision_for_soda_bottles.git

2. Navigate to the Project Directory:

cd comp_vision_for_soda_bottles/app

3. Set Up the Virtual Environment (Optional but recommended):

• For Windows:

  python -m venv env
  env\Scripts\activate
  

• For macOS and Linux:

  python3 -m venv env
  source env/bin/activate
  

4. Install Dependencies:

pip install -r requirements.txt

5. Run the Application:

uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Replace app.main:app with the appropriate module path if your FastAPI app instance is defined elsewhere.

6. Access the Web Service: Open your browser and visit http://localhost:8000 to interact with the application.

Note:

• The --reload flag in the uvicorn command enables automatic reloading of the server when changes are detected in the code. It is useful for development but should be omitted in a production environment.

How to Use:

• Home Page: Upload a photo to see the model's prediction.

• Results: Displays predictions with recognized soda bottle types.

• More Info: Insights on recognized classes and prediction statistics.

Make It Your Own:

Tweak, expand, and use this project as a base for further computer vision and transfer learning experiments.

Please Note:

For development only. Check conv_net_comparison.ipynb for model insights.