Coffee Level Detection

Coffee Level Detection is a computer vision project for detecting and counting coffee levels in images using manual annotation and deep learning. The project is modular, with a clear data pipeline from raw images to trained models and inference.

Features

• Interactive annotation tool for coffee level labeling
• Automated image splitting and augmentation
• Deep learning model training (coffeeCNN)
• Inference and prediction modules
• Structured annotation and summary files

Directory Structure

coffee_level_detection/
    dataset_collection/   # Annotation and dataset preparation
    preprosessing/        # Data preprocessing utilities
    training/             # Model training pipeline
    inference/            # Model inference and prediction
manual_levels/            # Coffee level annotation JSON files
masked_images/            # Masked coffee images
processed_images/         # Augmented images

Installation

1. Clone the repository:

   git clone https://github.com/PlayerPlanet/coffee_level_detection.git

2. Install dependencies with Poetry:

   poetry install

Usage

Annotate Coffee Levels

python -m coffee_level_detection.dataset_collection.labeler --samples 20

Process Images

python -m coffee_level_detection.dataset_collection.image_splitter --input_dir ./masked_images --output_dir ./processed_images

Train Model

python -m coffee_level_detection.training.train_coffeeCNN --f compiled_coffee_level_annotations.json --batch 10 --epochs 100 --img_dir processed_images

Inference

python -m coffee_level_detection.inference.predict --model_path <model.pt> --img_dir <images>

Data Format

• Annotation files: manual_levels/level_*.json
• Summary: coffee_level_annotations_summary.json
• Processed images: processed_images/

Development Standards

• All functions and classes must have docstrings
• Professional documentation and code comments
• Modular, maintainable codebase

License

MIT License