Lane Curvature Detection with BezierLaneNet

A deep learning-based lane detection and curvature estimation system that leverages Bezier curve representations for accurate and robust lane prediction. The project combines research experiments on lane curvature detection with a full PyTorch-based implementation (BezierLaneNet), providing both Jupyter notebook explorations and deployable training/inference pipelines.

---

✨ Features

• Lane detection using Bezier curve regression
• Lane curvature estimation and metrics from research notebooks
• Backbone models: ResNet, Custom ResNet
• Custom DSD loss function for better curve fitting
• Training and inference pipelines included (train.py, inference.py)
• Visualization tools for datasets, predictions, and curvature metrics
• Modular code structure (models, losses, utils)

---

⚙️ Installation

1. Clone the repository:

git clone https://github.com/<your-username>/Lane-Curvature-Detection-BezierLaneNet.git
cd Lane-Curvature-Detection-BezierLaneNet

2. Create a Python environment and install dependencies:

pip install -r requirements.txt

---

🚀 Usage

Training (BezierLaneNet)

python train.py --dataset <path_to_dataset> --epochs 50 --batch_size 16

Inference (BezierLaneNet)

python inference.py --image <path_to_image> --weights <path_to_weights>

📊 Dataset

• Supports lane detection datasets like TuSimple
• Utilities in utils/dataloader.py for dataset loading
• Custom visualization scripts in utils/visualize_dataset.py