Awesome

Sketch and Refine: Towards Fast and Accurate Lane Detection

Install

git clone https://github.com/passerer/SRLane.git
cd SRLane
conda create -n py38 python=3.8 -y # Create a new Python environment, optional.
conda activate py38 
pip install -r requirements.txt
pip install torch==1.13.0+cu117 # Install pytorch, modifying the CUDA version accordingly.
python setup.py develop

DATASET

Download CULane. Then modify dataset_path in configs/datasets/culane.py accordingly.

Train

Here is an example

CUDA_VISIBLE_DEVICES=0 python tools/main.py configs/exp_srlane_culane.py

Test

Performance

Here is an example:

CUDA_VISIBLE_DEVICES=0 python tools/main.py configs/exp_srlane_culane.py --load_from checkpoint/baseline.pth --validate

The results should be:

Runtime

Here is an example:

CUDA_VISIBLE_DEVICES=0 python tools/analysis_tools/speed_measure.py configs/exp_srlane_culane.py

Acknowledgement

• Turoad/CLRNet
• hirotomusiker/CLRerNet