🍽️ Food Image Classifier

This project trains a deep learning model to classify food images into one of 101 categories using the Food-101 dataset and ResNet18.

---

📁 Project Structure

food-image-classifier/
├── data/
│   └── food-101/
│       ├── images/               # Contains all original images
│       ├── images_split/         # Will contain train/ and test/ after splitting
│       └── meta/                 # Contains train.txt and test.txt split info
├── models/
│   ├── checkpoints/              # Contains checkpoint subdirs by config
│   └── final/                    # Stores final trained models
├── logs/                         # Contains training logs and summaries
├── output/                       # Stores predictions from predict.py
├── analysis/                     # Stores evaluation reports and plots
├── utils/
│   ├── data_loader.py
│   ├── logger.py
│   ├── model_builder.py
│   ├── trainer.py
│   └── validator.py
├── tests/
│   ├── __init__.py
│   ├── test_config.py            # Unit testing Configs
│   ├── test_utils.py             # Unit testing Utils
│   ├── test_data_loader.py       # Unit testing for data loading
│   ├── test_logger.py            # Unit testing for logging logic
│   └── test_predict.py           # Unit testing for predictions
├── config.py                     # CLI config parser
├── split_dataset.py              # Script to create train/test folders
├── train.py                      # Main training entrypoint
├── predict.py                    # Inference script
├── analysis/
│   └── prediction_analysis.py    # Model evaluation & visualization
├── app.py *(to be added)*       
├── requirements.txt
└── README.md

---

🛠️ Setup Instructions

✅ 1. Clone the Repository

git clone https://github.com/wafaahs/food-classifier.git
cd food-classifier

✅ 2. Set Up a Virtual Environment with Python 3.10 or 3.11

Make sure you are not using Python 3.13+, as PyTorch doesn't support it yet.

python3.10 -m venv venv
source venv/bin/activate

✅ 3. Install Dependencies

pip install --upgrade pip
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
pip install streamlit

Or use the requirements.txt:

torch==2.2.2
torchvision==0.17.2
torchaudio==2.2.2
numpy==1.26.4
pillow==10.3.0
tqdm==4.66.4
streamlit==1.35.0
matplotlib==3.8.4
seaborn==0.13.2
scikit-learn==1.5.0

pip install -r requirements.txt

---

📦 Download the Dataset

Run the following:

cd data/
wget https://data.vision.ee.ethz.ch/cvl/food-101.tar.gz
tar -xvzf food-101.tar.gz
rm food-101.tar.gz
cd ..

Make sure your dataset is structured like:

data/food-101/images/train/apple_pie/image1.jpg

You can verify category folders:

ls data/food-101/images/train | wc -l  # should show 101

If the train/ and test/ folders do not exist, run the provided helper script:

🧩 Split the Dataset into Train/Test Folders

python split_dataset.py

This will read from meta/train.txt and meta/test.txt and create:

data/food-101/images_split/train/
                         /test/

Then update DATA_DIR in train.py to point to images_split:

DATA_DIR = "data/food-101/images_split"

---

🚀 Train the Model

python train.py --epochs 10 --batch_size 64 --lr 0.0005 --optimizer sgd

🔧 Command-Line Options

You can pass the following optional arguments to customize training:

• --epochs: number of training epochs (default: 5)
• --batch_size: mini-batch size (default: 32)
• --lr: learning rate (default: 0.001)
• --optimizer: optimization algorithm (adam or sgd, default: adam)

If omitted, the script will fall back to defaults defined in config.py.

• Resumes from last checkpoint if available
• Logs progress to console and logs/
• Saves final model to models/final/
• Outputs epoch summaries in CSV format

---

🗓️ Output Structure

• Logs: logs/training_<timestamp>.log
• Summary CSV: logs/summary_<timestamp>.csv
• Checkpoints: models/checkpoints/e{epochs}_b{batch}_opt_lr.../
• Final model: models/final/e10_b64_sgd_lr0p0005_resnet.pth

---

🧠 Inference: Make Predictions

Once training is complete, use predict.py to run inference on the test dataset.

python predict.py \
  --model models/final/e10_b64_adam_lr0p001_resnet.pth \
  --data_dir data/food-101/images_split/test \
  --output_dir output/

🔧 Command-Line Options

• --model: path to the trained .pth model
• --data_dir: directory containing test images
• --output_dir: base folder for saving predictions

This will:

• Predict all images in the test directory
• Create a subdirectory in output/ named after the model config
• Save predictions.csv with the following format:

file_path, prediction, ground_truth
data/food-101/images_split/test/ramen/123.jpg, ramen, ramen
data/food-101/images_split/test/apple_pie/456.jpg, cheesecake, apple_pie

⚠️ Ensure the ground-truth class is inferable from the directory name for evaluation.

---

📊 Analyze Predictions

Use prediction_analysis.py to evaluate your model’s performance:

python analysis/prediction_analysis.py --subdir e10_b64_adam_lr0p001

🔧 Command-Line Options

• --subdir: name of the model subdirectory (e.g. e10_b64_adam_lr0p001) inside output/

This will automatically:

• Load output/e10_b64_adam_lr0p001/predictions.csv
• Create results in: analysis/e10_b64_adam_lr0p001/
• Generate:
  • classification_report.csv
  • summary.txt (with worst 5 classes)
  • confusion_matrix.png (multi-class)
  • class_metrics_bar_chart.png (precision/recall/F1)
  • Per-class 2×2 confusion matrices labeled with TP/FN/FP/TN and F1

---

🤪 Coming Next

• app.py with Streamlit UI
• GitHub-ready visual demo
• End-to-end notebook

Stay tuned!