LaneATT provides a state-of-the-art lane detection model for real-time applications, addressing the challenge of accurately identifying lane markings in diverse driving scenarios. It is designed for researchers and engineers in autonomous driving and advanced driver-assistance systems (ADAS).

How It Works

LaneATT employs an attention-guided approach to focus on relevant lane features, improving accuracy and efficiency. It utilizes a ResNet backbone for feature extraction and a novel attention mechanism to selectively process spatial information, leading to more robust lane predictions.

Quick Start & Requirements

• Install:

  conda create -n laneatt python=3.8 -y
  conda activate laneatt
  conda install pytorch==1.6 torchvision -c pytorch
  pip install -r requirements.txt
  cd lib/nms; python setup.py install; cd -
  

• Prerequisites: Python >= 3.5, PyTorch == 1.6, CUDA 10.2 (for NMS compilation), other dependencies in requirements.txt.
• Resources: Pre-trained models are available for download (1.3 GB).
• Docs: DATASETS.md for dataset setup.

Highlighted Details

• Achieves high F1 scores on CULane, TuSimple, and LLAMAS datasets.
• Offers real-time performance with FPS up to 250 on ResNet-18.
• Includes official metric implementations for CULane, TuSimple, and LLAMAS.
• Code structure is well-organized into models, datasets, and utilities.

Maintenance & Community

The project is associated with the CVPR 2021 paper "Keep your Eyes on the Lane: Real-time Attention-guided Lane Detection." No specific community channels or active maintenance signals are provided in the README.

Licensing & Compatibility

The repository does not explicitly state a license. The code is provided for research purposes related to the CVPR 2021 paper.

Limitations & Caveats

The provided PyTorch 1.6 and CUDA 10.2 versions are specific; using other versions may lead to slightly different metric results. The unofficial CULane metric implementation is faster but has minor deviations from the official C++ version.