Ultra-Fast-Lane-Detection

The implementation of the paper "Ultra Fast Structure-aware Deep Lane Detection".

The evaluation code is modified from SCNN and Tusimple Benchmark.

Demo

Install

a. Clone the project

git clone https://github.com/cfzd/Ultra-Fast-Lane-Detection
cd Ultra-Fast-Lane-Detection

b. Create a conda virtual environment and activate it

conda create -n lane-det python=3.7 -y
conda activate lane-det

c. Install dependencies

# If you don't have pytorch
conda install pytorch torchvision cudatoolkit=10.1 -c pytorch 

pip install -r requirements.txt

d. Install CULane evaluation tools. This tools requires OpenCV C++. Please follow here to install OpenCV C++. When you build OpenCV, remove the paths of anaconda from PATH or it will be failed.

# First you need to install OpenCV C++.
cd evaluation/culane
make

e. Data preparation

Download CULane and Tusimple. Then extract them to $CULANEROOT and $TUSIMPLEROOT. For Tusimple, the segmentation annotation is not provided, hence we need to generate segmentation from the json annotation.

cd $TUSIMPLEROOT
python scripts/convert_tusimple.py --root $TUSIMPLEROOT
# this will generate segmentations and 2 list files train_gt.txt test.txt

Get Started

Copy a config from configs/culane.py or configs/tusimple.py, then modifiy data_root, log_path and other settings in your config.

For single gpu training, run

python train.py configs/path_to_your_config

For multi-gpu training, run

sh launch_training.sh

# If there is no pretrained torchvision model, multi-gpu training may result in multiple downloading. You can first download the corresponding models manually, and then restart the multi-gpu training.

python -m torch.distributed.launch --nproc_per_node=$NGPUS train.py configs/path_to_your_config

Besides config style settings, we also support command line style one. You can override a setting like

python train.py configs/path_to_your_config --batch_size 8

The batch_size will be set to 8 during training.

To visualize the log with tensorboard, run

tensorboard --logdir log_path --bind_all

Trained models

We provide two trained Res-18 models on CULane and Tusimple.

Dataset	Metric paper	Metric This repo	Avg FPS on GTX 1080Ti	Model
Tusimple	95.87	95.82	306	GoogleDrive
CULane	68.4	69.7	324	GoogleDrive

For evaluation, run

mkdir tmp

python test.py configs/culane.py --test_model path_to_culane_18.pth --test_work_dir ./tmp

python test.py configs/tusimple.py --test_model path_to_tusimple_18.pth --test_work_dir ./tmp

Same as training, multi-gpu evaluation is also supported.

Citation