Fruit to Wine Tutorial

A beginner-friendly machine learning project that predicts the type of fruit from an image and recommends a wine-making tutorial based on the result.

This project was built to learn machine learning by building small, understandable systems rather than optimizing for production-level accuracy.

---

What it does

1. Upload an image of a fruit
2. A trained image classification model predicts the fruit
3. The app recommends a curated YouTube tutorial for making wine from that fruit

The entire flow is presented through a simple Streamlit web app.

---

Tech stack

• Python
• PyTorch
• Torchvision (ResNet18 with transfer learning)
• Streamlit
• Pillow (PIL)

---

Project structure

wine-classification/
│
├── app.py                  # Streamlit entry point
│
├── src/                     # Core logic
│   ├── train_model.py       # Model training
│   ├── predict_fruit.py     # Inference logic
│   ├── label_mapping.py     # Dataset to label mapping
│   └── fruit_to_tutorial.py # Fruit to wine tutorial mapping
│
├── models/
│   └── fruit_model.pth      # Trained model weights
│
├── data/
│   └── dataset/             # ImageFolder-style dataset
│
├── LICENSE
└── README.md

---

How to run

1. Install dependencies

pip install torch torchvision streamlit pillow

2. Run the app

streamlit run app.py

---

Model notes

• Uses transfer learning with a pretrained ResNet18
• Trained on a small custom dataset
• Confidence scores typically range between 55–75 percent, which is expected for a beginner-scale model
• The goal is learning and experimentation, not production accuracy

---

Why this project exists

This project was created in response to requests for a simple and honest beginner machine learning example.

Design goals:

• Minimal abstraction
• Clear separation of training, inference, and application logic
• Easy to read, modify, and extend

---

Possible improvements

• Add more training data
• Display prediction confidence in the UI
• Normalize inputs using ImageNet statistics
• Apply data augmentation during training

---

License

MIT