Lane Detection and Classification using Cascaded CNNs

This repository contains inference code for Lane Detection and Classification using a cascaded approach with Convolutional Neural Networks (CNNs). It utilizes ERFNet for lane segmentation and LCNet for lane classification.

📌 Overview

The system performs two main tasks:

1. Lane Segmentation: Identifies the pixels belonging to lanes in an image using ERFNet (Efficient Residual Factorized Network).
2. Lane Classification: Classifies the detected lanes into different types (e.g., continuous, dashed) using LCNet.

This project is designed for inference using pretrained models.

📂 Project Structure

├── main.ipynb           # Main Jupyter Notebook for inference and visualization
├── models/              # Neural network model definitions
│   ├── erfnet.py        # ERFNet architecture
│   └── lcnet.py         # LCNet architecture
├── pretrained/          # Directory for pretrained model weights
│   ├── erfnet_tusimple.pth
│   └── classification_64_3class.pth
├── images/              # Directory for input images
├── functions.py         # Helper functions for visualization
├── loss.py              # Loss functions (used during training)
└── requirements.txt     # Python dependencies

🛠️ Prerequisites

• Python 3.x
• PyTorch
• Torchvision
• Numpy
• Matplotlib
• Pillow (PIL)

⚙️ Installation

1. Clone the repository (if you haven't already):

   git clone <repository-url>
   cd Cascade-LD-master

2. Install dependencies:

   pip install -r requirements.txt

🚀 Usage

1. Open the Jupyter Notebook:

   jupyter notebook main.ipynb

2. Run the cells:

   • The notebook will load the pretrained models (erfnet_tusimple.pth and classification_64_3class.pth).
   • It will load a sample image from the images/ directory.
   • It performs lane segmentation and classification.
   • Finally, it visualizes the results, showing the detected lanes colored by their instance and class.

📊 Models & Dataset

• Segmentation Model: ERFNet trained on the TuSimple dataset.
• Classification Model: LCNet trained to classify lane types (Continuous, Dashed, Double-Dashed).

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📜 License

This project is open source and available under the MIT License.