🩺 End-to-End Deep Learning Project – Skin Condition Multi-Label Classifier

This project builds a multi-label skin-condition classifier using a VGG16 pretrained model, frozen CNN layers, and a custom ANN classification head. The training pipeline is fully modular with YAML-based configs, automated data ingestion, model preparation, training, and prediction pipelines.The final model achieves strong accuracy for all 10 classes and is deployed through a Flask web app that performs real-time image prediction.Model achieved a training binary accuracy of 91.70% and a validation binary accuracy of 93.99% in 1 epoch only, demonstrating strong multi-label performance across all 10 skin-condition categories.

📦 Dataset

Google Drive Download Link:
https://drive.google.com/file/d/1XaNxpHP3XwDyKjEw-1wirLcgLqMRSsV-/view?usp=sharing

---

✅ Workflows Completed

1. Update config.yaml
2. Update params.yaml
3. Create entity classes (config_entity.py & artifact_entity.py)
4. Implement Configuration Manager (configuration.py)
5. Build Data Ingestion Component
6. Build Base Model Preparation Component
7. Build Model Training Component
8. Build Training Pipeline
9. Build Prediction Pipeline
10. Build Flask App + HTML UI
11. Run end-to-end system successfully

---

🚀 How to Run the Project

1️⃣ Clone the Repository

git clone <repo-link>
cd <project-folder>

2️⃣ Create & Activate Conda Environment

conda create -n cnn python=3.10 -y
conda activate cnn

3️⃣ Install Requirements

pip install -r requirements.txt

4️⃣ Run the Full Training Pipeline

python main.py

5️⃣ Run the Flask App

python app.py

Then open your browser

http://127.0.0.1:5000/predict

🧠 Model Details

• Model: VGG16 pretrained on ImageNet
• include_top: False
• Frozen Layers: All convolutional layers
• Custom Head:
  • Dense → ReLU
  • Dropout
  • Dense → Sigmoid (multi-label output)
• Loss Function: Binary Cross-Entropy
• Optimizer: Adam
• Learning rate: from params.yaml
• Classes: 10 skin conditions

🖥️ Web App Preview

A simple Flask -based UI that supports:

• Uploading an image
• Displaying the uploaded image
• Showing predictions for each skin condition:
  • Probability values
  • "Yes/No" prediction
  • Colour-coded table (Green = Yes, Red = No)

Ideal for demonstrations, testing, or local deployment.

🛠️ Technologies Used

• Python
• TensorFlow / Keras
• Pandas / NumPy
• OpenCV / Pillow
• Flask
• YAML (configuration-driven pipeline)
• Modular MLOps-style folder structure