🍕 FoodVision Mini – Transfer Learning with EfficientNet

📌 Overview

This project builds an image classification model to identify food items (Pizza, Steak, Sushi) using Transfer Learning with a pre-trained EfficientNet-B0 model.

Instead of training from scratch, we reuse knowledge from ImageNet to achieve high accuracy with limited data and faster training.

---

🎯 Problem Statement

Classify food images into:

• Pizza 🍕
• Steak 🥩
• Sushi 🍣

---

🧠 Transfer Learning

• Use pretrained EfficientNet-B0
• Freeze feature extraction layers
• Replace classifier head for 3 classes

---

🏗️ Model Architecture

model.classifier = nn.Sequential(
    nn.Dropout(p=0.2),
    nn.Linear(in_features=1280, out_features=3)
)

---

⚙️ Workflow

1. Data preprocessing (resize, normalize, augment)
2. Load pretrained model
3. Freeze backbone layers
4. Train classifier
5. Evaluate on validation + custom images

---

📊 Results

• Good performance on unseen images
• Example predictions:
  • Pizza: ~49%
  • Steak: ~53%
  • Sushi: ~81%

---

📂 Project Structure

FoodVision_Mini_Transfer_Learning.ipynb
data/
models/
README.md

---

▶️ How to Run

pip install torch torchvision

Run notebook in Google Colab: FoodVision_Mini_Transfer_Learning.ipynb

---

💡 Key Learnings

• Transfer learning reduces training time
• EfficientNet provides strong feature extraction
• Works well with small datasets

---

🚀 Future Work

• Add more classes
• Fine-tune full model
• Deploy using Flask/FastAPI

---

🛠️ Tech Stack

• Python
• PyTorch
• Torchvision
• Google Colab