This repo is the PyTorch implementation for LaneSegNet:
We propose LaneSegNet, a robust lane detection method that introduces a progressive lane reconstruction pipeline from lane points to lane segments, ultimately forming complete lanes. Specifically, lane points and multiple reference points are initially extracted and represented as hierarchical graph nodes. A graph enhancement module is utilized to capture both global and local structural relationships between reference points and lane points, respectively. Then a lane segment is constructed by associating adjacent lane points with each reference point. Multiple lane segments will be utilized to reconstruct a complete lane. LaneSegNet can efficiently establish the topological relationships of sparse lane points across different spatial scales and robustly detect lanes in challenging situations.
- Create a conda virtual environment and activate it.
conda create -n lanesegnet python=3.7 -y conda activate lanesegnet conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.1 -c pytorch -y pip install -r requirements/build.txt
- Clone this repository and enter it:
git clone https://github.com/ustclbh/LaneSegNet.git cd LaneSegNet python setup.py develop
Download datasets and put it into [data-path] folder. And edit the data_root in the config file to your dataset path.
The directory structure should be like follows:
[data-path]/culane
├── driver_23_30frame
├── driver_37_30frame
├── driver_100_30frame
├── driver_161_90frame
├── driver_182_30frame
├── driver_193_90frame
└── list
└── test_split
| ├── test0_normal.txt
| ├── test1_crowd.txt
| ├── test2_hlight.txt
| ├── test3_shadow.txt
| ├── test4_noline.txt
| ├── test5_arrow.txt
| ├── test6_curve.txt
| ├── test7_cross.txt
| └── test8_night.txt
└── train.txt
└── train_gt.txt
└── test.txt
└── test_gt.txt
└── val.txt
The directory structure should be like follows:
[data-path]/tusimple
├── clips
├── label_data_0313.json
├── label_data_0531.json
├── label_data_0601.json
├── label_data_0601.json
├── test_label.json
└── test_baseline.json
To train the model, run the following commands.
cd tools
sh dist_train.sh culane final_exp_res18_s8 ./output