CLRNet provides a PyTorch implementation for cross-layer refinement network-based lane detection, addressing the need for accurate lane localization by leveraging contextual information and local lane features. It is targeted at researchers and practitioners in autonomous driving and computer vision.

How It Works

CLRNet employs a cross-layer refinement strategy to fuse multi-scale features, enhancing both contextual understanding and precise localization of lane markings. This approach aims to improve accuracy by integrating detailed local features with broader contextual cues, leading to state-of-the-art performance.

Quick Start & Requirements

• Install: Clone the repository, create a conda environment (optional), and install dependencies via python setup.py build develop. PyTorch installation is recommended via conda (conda install pytorch torchvision cudatoolkit=10.1 -c pytorch) or pip (pip install torch==1.8.0 torchvision==0.9.0).
• Prerequisites: Ubuntu 18.04/20.04, Python >= 3.8, PyTorch >= 1.6, CUDA (tested with 10.2).
• Data: Download CULane, Tusimple, or LLAMAS datasets and create symbolic links to the $CLRNET_ROOT/data directory. Tusimple requires an additional script to generate segmentation masks.
• Links: Papers With Code

Highlighted Details

• Achieves state-of-the-art results on CULane, Tusimple, and LLAMAS datasets.
• Offers multiple backbone options including ResNet-18, ResNet-34, ResNet-101, and DLA-34.
• Supports visualization of detection results during testing with the --view flag.
• Provides detailed performance metrics (mF1, F1@50, F1@75, Acc, FDR, FNR) for various datasets and backbones.

Maintenance & Community

The project is associated with the CVPR 2022 paper "CLRNet: Cross Layer Refinement Network for Lane Detection." It acknowledges contributions from several open-source projects including mmdetection, pytorch/vision, and Ultra-Fast-Lane-Detection.

Licensing & Compatibility

The repository does not explicitly state a license in the README.

Limitations & Caveats

The installation and testing were primarily performed on Ubuntu 18.04 and 20.04 with specific versions of PyTorch and CUDA. The Tusimple dataset requires a custom script to generate segmentation masks, as they are not provided directly.