PyTorch implementation of the paper "Oblique Convolution: A Novel Convolution Idea for Redefining Lane Detection"
This paper is published at IEEE Transactions on Intelligent Vehicles. Link
- [2023-02-22] Release the initial code for Oblique Convolution.
feature map trained with Tusimple
ORM architecture. α1, α2, α3, α4, -α1, -α2, -α3, -α4 is the angle obtained by clustering. Group convolution is used to separate them between different angles and finally stitch them together. Through rotating different angles and grouping convolution, the features of all vertical lane lines are obtained.
SSAM first performs the maxpooling, and then uses the strip 3 × H deformable convolution with a certain width to directly extract the information of the whole lane and give different pixel points importance.
The visualization results compared with other methods, where red and green represent false positive and false negative respectively.
Environment configuration: Clone repo and install requirements.txt in a Python>=3.6.0 environment, including PyTorch>=1.7.
git clone https://github.com/gongyan1/Oblique-Convolution.git
pip install -r requirements.txt # install
please modify data_root and log_path in your configs/culane.py or configs/tusimple.py config according to your environment.
data_root is the path of your CULane dataset or Tusimple dataset. log_path is where tensorboard logs, trained models and code backup are stored.
First, change some some hyperparameters in configs/path_to_your_config
python train.py configs/path_to_your_config
python test.py configs/culane.py --test_model path_to_culane.pth --test_work_dir ./tmp
python test.py configs/tusimple.py --test_model path_to_tusimple.pth --test_work_dir ./tmp
Performance comparison with other lane detection methods on Tusimple dataset.
Performance comparison with other lane detection methods on CULane dataset.
@article{zhang2023oblique, title={Oblique Convolution: A Novel Convolution Idea for Redefining Lane Detection}, author={Zhang, Xinyu and Gong, Yan and Lu, Jianli and Li, Zhiwei and Li, Shixiang and Wang, Shu and Liu, Wenzhuo and Wang, Li and Li, Jun}, journal={IEEE Transactions on Intelligent Vehicles}, year={2023}, publisher={IEEE} }