🚗 D3-CNN: Driver Distraction Detector

D3-CNN is a simple Convolutional Neural Network (CNN) built from scratch using NumPy to detect driver distractions from images and videos. It’s an academic project designed to deeply understand how CNNs work internally — with no deep learning frameworks like PyTorch or TensorFlow used.

It is trained purely on CPU, and also features a Streamlit-based frontend for easy testing with image or video input.

---

📸 Sample Input vs Prediction

Input Image	Prediction
	Texting

---

🧠 Model Architecture

The CNN is simple and built from scratch using NumPy only:

self.layers = [
    Conv2D(filters=32, kernel_size=3, in_channels=3, padding=1),
    ReLU(),
    MaxPool2D(),
    Conv2D(filters=64, kernel_size=3, in_channels=32, padding=1),
    ReLU(),
    MaxPool2D(),
    Flatten(),
    Dense(input_size=16*16*64, output_size=128),
    ReLU(),
    Dense(input_size=128, output_size=NUM_CLASSES),
    Softmax()
]

📊 Architecture Diagram

Note: Please include the architecture diagram as architecture.png in the project root directory.

---

📂 Dataset

• A mix of publicly available data and self-collected images.
• Download the dataset from this Google Drive link: 📎 Download Dataset

---

✅ Model Performance

• Trained on CPU from scratch
• Test Accuracy: ~75%
• Lightweight model for quick experimentation and learning

---

🚀 Setup & Installation

1. Clone the repository

git clone https://github.com/AS-0167/D3-CNN.git
cd D3-CNN

2. Install dependencies

pip install -r requirements.txt

---

🧪 Running the Streamlit App

You can upload an image or a video and get distraction predictions.

streamlit run app.py

---

🏋️‍♂️ Training the Model (Optional)

To retrain the model from scratch using the dataset:

1. Ensure dependencies are installed:

   pip install -r requirements.txt

2. Run the training notebook:

   jupyter notebook train-detect-distraction.ipynb

---

🎓 About the Project

This project was created as a hands-on academic experiment to:

• Learn how CNNs work internally
• Understand forward and backward propagation using NumPy
• Handle dataset collection and preprocessing
• Build a mini deep learning pipeline without relying on big frameworks

No libraries like TensorFlow or PyTorch were used.

---

👨‍💻 Author

[AS-0167] 🧑‍🎓 Student, Enthusiastic Learner of Machine Learning & Computer Vision

---