HWLane

PyTorch implementation of the paper "[HWLane: HW-Transformer for Lane Detection]".

CULane demo

1.-.Trim.-.Trim.-.Trim.-.Trim.-.Trim.mp4

Introduction

HWLane achieves SOTA results on VIL-100, CULane, and Tusimple Dataset.

Get started

Clone the HWLane repository
```
git clone https://github.com/Cuibaby/HWLane.git
```
We call this directory as $HWLane_ROOT
Create a conda virtual environment and activate it (conda is optional)
```
conda create -n HWLane python=3.8 -y
conda activate HWLane
```

Install dependencies

# Install pytorch firstly, the cudatoolkit version should be same in your system. (you can also use pip to install pytorch and torchvision)
conda install pytorch torchvision cudatoolkit=10.1 -c pytorch

# Or you can install via pip
pip install torch torchvision

# Install python packages
pip install -r requirements.txt

Data preparation

Download VIL100, CULane and Tusimple. Then extract them to $VIL100ROOT $CULANEROOT and $TUSIMPLEROOT. Create link to data directory.

cd $MFIALane_ROOT
mkdir -p data
ln -s $VIL100ROOT data/VIL100
ln -s $CULANEROOT data/CULane
ln -s $TUSIMPLEROOT data/tusimple

For CULane, you should have structure like this:

$CULANEROOT/driver_xx_xxframe    # data folders x6
$CULANEROOT/laneseg_label_w16    # lane segmentation labels
$CULANEROOT/list                 # data lists

For Tusimple, you should have structure like this:

$TUSIMPLEROOT/clips # data folders
$TUSIMPLEROOT/lable_data_xxxx.json # label json file x4
$TUSIMPLEROOT/test_tasks_0627.json # test tasks json file
$TUSIMPLEROOT/test_label.json # test label json file

For Tusimple, the segmentation annotation is not provided, hence we need to generate segmentation from the json annotation.

python tools/generate_seg_tusimple.py --root $TUSIMPLEROOT
# this will generate seg_label directory

For VIL100, you should have structure like this:

$VIL100ROOT/Annotations 
$VIL100ROOT/data 
$VIL100ROOT/JPEGImages
$VIL100ROOT/Json
$VIL100ROOT/list
$VIL100ROOT/test

Install CULane evaluation tools.

This tools requires OpenCV C++. Please follow here to install OpenCV C++. Or just install opencv with command sudo apt-get install libopencv-dev

Then compile the evaluation tool of CULane.
```
cd $HWLane_ROOT/runner/evaluator/culane/lane_evaluation
make
cd -
```
Note that, the default opencv version is 3. If you use opencv2, please modify the OPENCV_VERSION := 3 to OPENCV_VERSION := 2 in the Makefile.

If you have problems installing the C++ version, you can remove the $lane_evaluation and change the 'type=Py_CULane' in the config file to use the pure Python version for evaluation.

Training

For training, run

CUDA_VISIBLE_DEVICES=0,1,2,3,4 python main.py [configs/path_to_your_config] --gpus [gpu_ids]

For example, run

CUDA_VISIBLE_DEVICES=0,1,2,3,4 python main.py configs/culane.py --gpus 0 1 2 3

Testing

For testing, run

CUDA_VISIBLE_DEVICES=0,1,2,3,4 python main.py c[configs/path_to_your_config] --validate --load_from [path_to_your_model] [gpu_num]

For example, run

CUDA_VISIBLE_DEVICES=0,1,2,3,4 python main.py configs/culane.py --validate --load_from culane.pth --gpus 0 1 2 3

CUDA_VISIBLE_DEVICES=0,1,2,3,4 python main.py configs/tusimple.py --validate --load_from tusimple.pth --gpus 0 1 2 3

CUDA_VISIBLE_DEVICES=0,1,2,3,4 python main.py configs/vilane.py --validate --load_from vilane.pth --gpus 0 1 2 3

We provide three trained ResNet/VGG models on VIL100, CULane and Tusimple.

Dataset	Backbone	Metric paper	Metric This repo	Model
VIL100	ResNet34	91.9	91.9	Comming Soon
Tusimple	ResNet18	96.83	96.83	Comming Soon
CULane	VGG16	76.9	76.9	Comming Soon

Visualization

Just add --view.

For example:

python main.py configs/culane.py --validate --load_from culane.pth --gpus 0 1 2 3 --view

You will get the result in the directory: work_dirs/[DATASET]/xxx/vis.

Citation

@ARTICLE{10507728,
  author={Zhao, Jing and Qiu, Zengyu and Hu, Huiqin and Sun, Shiliang},
  journal={IEEE Transactions on Intelligent Transportation Systems}, 
  title={HWLane: HW-Transformer for Lane Detection}, 
  year={2024},
  volume={},
  number={},
  pages={1-11},
  keywords={Lane detection;Feature extraction;Computational modeling;Task analysis;Visualization;Current transformers;Convolutional neural networks;Deep learning;lane detection;transformer;self-knowledge distillation},
  doi={10.1109/TITS.2024.3386531}}


@ARTICLE{9872124,  
author={Qiu, Zengyu and Zhao, Jing and Sun, Shiliang},  
journal={IEEE Transactions on Intelligent Transportation Systems},   
title={MFIALane: Multiscale Feature Information Aggregator Network for Lane Detection},   
year={2022},  
volume={},  
number={},  
pages={1-13},  
doi={10.1109/TITS.2022.3195742}
}

Thanks

The code is modified from RESA and SCNN, Tusimple Benchmark. It's also recommended for you to try LaneDet.

Cuibaby/HWLane