DriverNet

Distracted Driver Detection Using Deep Learning Through Depth-Grouped Pooling and Parallel Multi-Aspect Patching

---

0. Prerequisites

• Python 3.10+
• uv: Install UV
• GPU(CUDA) available environment

1. Create Venv & Install Dependencies

uv venv
source .venv/bin/activate
uv sync

2. Download Dataset

Original Dataset was sourced from Kaggle.

Download preprocessed images from dohyeoplim/drivernet-images-depth-v2.

3. Configure Pipeline

Set data directories and model hyperparameters in config.yaml.

4. Run Training & Create Submission

uv run main.py --train-and-submit

---

Method 1

Depth Grouped Pooling (DGP)

---

Method 2

Parallel Multi-Aspect Patching (P-MAP)

---

Method 3

Strong Data Augmentations

---

Method 4

Post-Processing for Stability

---