AIH (AM I HUMAN) ?

"Are you a human, or have the aliens taken over?" 👽 Inspired by the Men in Black series, AIH is a fun and quirky project designed to determine if you're truly human or if you're just a cleverly disguised extraterrestrial. With a mix of modern AI techniques, this tool can assess not only your human status but also your age, gender, whether you're alive, or if you might just be a brilliant spoof. Don't worry, I've got you covered!

---

⚙️ Technologies Used

• Transfer Learning (MobileNetV3): Using the MobileNetV3 model for efficient image classification.
• FastAPI: Backend for handling requests and serving the model efficiently.
• DVC (Data Version Control): For managing model data and tracking changes.
• MLflow: To manage the entire machine learning lifecycle including experimentation, model tracking, and deployment.
• YOLO (You Only Look Once): For object detection, especially detecting faces to evaluate human-like features.
• Dagshub: Cloud-based model and data management system.
• Jinja: For dynamically generating HTML templates to serve results.
• Uvicorn: ASGI server for running the FastAPI app.

---

🧠 How It Works

1. Model Training: Fine-tune a MobileNetV3 model using a dataset of human images. The model is trained to classify various features like age, gender, and whether the input appears human.

2. FastAPI Backend: FastAPI is used to build a lightweight API that takes an image as input and returns predictions (Human Status, Age, Gender, etc.). It’s fast, lightweight, and super efficient.

3. Model Management in the Cloud:

   • The model is completely managed in the cloud via Dagshub.
   • Every time the server (FastAPI) restarts, the model is fetched from the cloud and loaded for prediction.
   • DVC is used for version control to manage and track data and model changes.
   • MLflow is used to track the machine learning lifecycle, from experimentation to deployment.

4. Prediction Process: When you upload an image, YOLO detects if the face is human, and the MobileNetV3 model classifies it into human categories. The result is served via a friendly UI built using Jinja templates.

---

🎉 Fun Features

• Human Detection: The AI will tell you if you're human or just pretending to be one!
• Age & Gender Prediction: Get an estimate of your age and gender based on the image you provide.
• Spoof Detection: Using the techniques to detect spoof faces or avatars.
• Interactive UI: Dynamic, fun, and interactive web interface built with Jinja templates.

---

💻 Installation and Setup

1. Clone the Repo

git clone https://github.com/RijoSLal/AIH.git
cd AIH

2. Install Dependencies

pip install -r requirements.txt

3. Run the FastAPI App

uvicorn service:app --reload

4. Start the App

Open your browser and go to http://127.0.0.1:8000/ to interact with the app!

---

⚙️ Training the Model

To train the model, follow the steps below:

1. Run the model.ipynb to train and fine-tune the MobileNetV3 model on your dataset.

   jupyter notebook model.ipynb

   The model will be trained using the transfer learning approach, leveraging the MobileNetV3 pre-trained weights.

2. Register the Model on the Server: Once training is complete, use the model_registration.ipynb to register the trained model in the server. This will save the model on Dagshub for cloud management.

   jupyter notebook model_registration.ipynb

   This script will upload the model to the cloud and register it for future use, ensuring that it's available whenever the server restarts.

---

🌩️ MLOps Workflow with Dagshub

The model and data are managed in the cloud via Dagshub, ensuring that:

1. Version Control: All models and datasets are versioned using DVC.
2. Tracking: The entire machine learning workflow, including experimentation, is tracked using MLflow.
3. Cloud-based Deployment: Whenever the FastAPI server restarts, the model is automatically fetched from Dagshub and deployed without any additional setup.

💡 Change Your Dagshub Repo Name Don’t forget to update the Dagshub remote URL in .dvc/config and .mlflow tracking if you fork this project or rename it:

Example:

dvc remote modify origin url https://dagshub.com/your-username/your-repo-name.git

🧪 Contributing

Since this is a solo project, contributions are welcome but not expected. However, if you think of any fun ideas, bug fixes, or improvements, feel free to fork the repo and submit a pull request!

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

Note:

No alien or spoofed face will get past my detectors! 😉