Skip to content

Latest commit

 

History

History
319 lines (277 loc) · 12.8 KB

ENGLISH.md

File metadata and controls

319 lines (277 loc) · 12.8 KB

PPlanedet: A Toolkit for lane detection based on PaddlePaddle

In this project, we develop a toolkit for lane detection to facilitate research. Welcome to join us to make this project more perfect and practical.

If you do not have enough compute resource, we recommend that you can run our project at AiStudio, which can provide you with V100(32GB memory) for free. We also opened source the chinese version at AiStudio. Project link is here

News

If you want to learn more changes, you can refer to History.

  • [2023-02-24] : We fixed some bugs in PPLanedet. CLRNet is still under fixing. If you want to achieve high performance, we recommend that you can try CondLaneNet.
  • [2023-02-24] 🔥 We released the version5 of the PPLanedet. In V5, we reproduced more backbones and necks like CSPRepBiFPN, which is used in YOLOv6. With aforementioned components, we achieved the state-of-the-art performance on CULane with CondLaneNet. Compared with vanilla CondLaneNet, Our CondLaneNet achieves 79.92 F1 score and only contains 11M parameters. More details you can find in CondLaneNet config.

Introduction

PPlanedet is developed for lane detection based on PaddlPaddle. PaddlePaddle is a high performance Deep learning framework. The idea behind the pplanedet is to facilitate researchers who use PaddlePaddle to do research about lane detection. If you have any suggestions about our project, you can contact me.

Overview

Models Components
Backbones
Necks
Losses
Metrics
  • Accuracy
  • FP
  • FN
Datasets
Data Augmentation
  • RandomLROffsetLABEL
  • Resize
  • RandomUDoffsetLABEL
  • RandomCrop
  • CenterCrop
  • RandomRotation
  • RandomBlur
  • Normalize
  • RandomHorizontalFlip
  • Colorjitters
  • RandomErasings
  • GaussianBlur
  • RandomGrayScale
  • Alaug

Installation

step 1 Install PaddlePaddle>=2.4.0(you can refer to official documentation)

conda create -n pplanedet python=3.8 -y
conda activate pplanedet
conda install paddlepaddle-gpu==2.4.1 cudatoolkit=10.2 --channel https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/Paddle/

step2 Git clone PPlanedet

git clone https://github.com/zkyseu/PPlanedet

step3 Install PPlanedet

cd PPlanedet
pip install -r requirements.txt
python setup.py build develop

Data preparation

CULane

Download CULane. Then extract them to $CULANEROOT. Create link to data directory.

cd $LANEDET_ROOT
mkdir -p data
ln -s $CULANEROOT data/CULane

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

Tusimple

Download Tusimple. Then extract them to $TUSIMPLEROOT. Create link to data directory.

cd $LANEDET_ROOT
mkdir -p data
ln -s $TUSIMPLEROOT data/tusimple

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
# python tools/generate_seg_tusimple.py --root /root/paddlejob/workspace/train_data/datasets --savedir /root/paddlejob/workspace/train_data/datasets/seg_label

Getting Started

Training

For training, run(shell scripts are under folder script). More training details are in documentation

# training on single-GPU
export CUDA_VISIBLE_DEVICES=0
python tools/train.py -c configs/scnn/resnet50_tusimple.py
# training on multi-GPU
export CUDA_VISIBLE_DEVICES=0,1,2,3
python -m paddle.distributed.launch tools/train.py -c configs/scnn/resnet50_tusimple.py

Testing

For testing, run

python tools/train.py -c configs/scnn/resnet50_tusimple.py \
                      --load /home/fyj/zky/tusimple/new/pplanedet/output_dir/resnet50_tusimple/latest.pd \
                      --evaluate-only

Inference/Demo

See tools/detect.py for detailed information.

python tools/detect.py --help

usage: detect.py [-h] [--img IMG] [--show] [--savedir SAVEDIR]
                 [--load_from LOAD_FROM]
                 config

positional arguments:
  config                The path of config file

optional arguments:
  -h, --help            show this help message and exit
  --img IMG             The path of the img (img file or img_folder), for
                        example: data/*.png
  --show                Whether to show the image
  --savedir SAVEDIR     The root of save directory
  --load_from LOAD_FROM
                        The path of model

To run inference on example images in ./images and save the visualization images in vis folder:

# first you should add 'seg = False' in your config 
python tools/detect.py configs/scnn/resnet50_tusimple.py --img images\
          --load_from model.pd --savedir ./vis

If you want to save the visualization of the segmentation results, you can run the following code

# first you should add 'seg = True' in your config 
python tools/detect.py configs/scnn/resnet50_tusimple.py --img images\
          --load_from model.pd --savedir ./vis

Speed test

If you want to test the inference speed, you can run the following code. It should be noted that test script is written by python instead of C++, there may exist some difference between official speed and our test speed.

 python tools/test_speed.py configs/condlane/cspresnet_50_culane.py --model_path output_dir/cspresnet_50_culane/model.pd

License

PPlanedet is released under the MIT license. We only allow you to use our project for academic uses.

Acknowledgement

  • Thanks PASSL for providing Hook codes
  • Thanks lanedet for providing model codes.

Citation

If you find our project useful in your research, please consider citing:

@misc{PPlanedet,
    title={PPlanedet, A Toolkit for lane detection based on PaddlePaddle},
    author={Kunyang Zhou},
    howpublished = {\url{https://github.com/zkyseu/PPlanedet}},
    year={2022}
}

model reproduced in our project

@Inproceedings{pan2018SCNN,  
  author = {Xingang Pan, Jianping Shi, Ping Luo, Xiaogang Wang, and Xiaoou Tang},  
  title = {Spatial As Deep: Spatial CNN for Traffic Scene Understanding},  
  booktitle = {AAAI Conference on Artificial Intelligence (AAAI)},  
  month = {February},  
  year = {2018}  
}

@InProceedings{qin2020ultra,
author = {Qin, Zequn and Wang, Huanyu and Li, Xi},
title = {Ultra Fast Structure-aware Deep Lane Detection},
booktitle = {The European Conference on Computer Vision (ECCV)},
year = {2020}
}

@article{2017ERFNet,
author = {E.Romera, J.M.Alvarez, L.M.Bergasa and R.Arroyo},
title = {ERFNet: Efficient Residual Factorized ConvNet for Real-time Semantic Segmentation},
journal = {IEEE Transactions on Intelligent Transportation Systems(T-ITS)},
year = {2017}
}

@InProceedings{2021RESA,
author = {Zheng, Tu and Fang, Hao and Zhang, Yi and Tang, Wenjian and Yang, Zheng and Liu, Haifeng and Cai, Deng},
title = {RESA: Recurrent Feature-Shift Aggregator for Lane Detection},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2021}
}

@InProceedings{DeepLabV3+,
author = {Chen, Liang-Chieh, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam},
title = {Encoder-decoder with atrous separable convolution for semantic image segmentation},
booktitle = {In Proceedings of the European conference on computer vision(ECCV)},
year = {2018}
}

@InProceedings{CondLaneNet,
author = {Liu, Lizhe and Chen, Xiaohao and Zhu, Siyu and Tan, Ping},
title = {CondLaneNet: a Top-to-down Lane Detection Framework Based on Conditional Convolution},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
year = {2021}
}

@article{RTFormer,
author = {Wang, Jian, Chenhui Gou, Qiman Wu, Haocheng Feng, Junyu Han, Errui Ding, and Jingdong Wang},
title = {RTFormer: Efficient Design for Real-Time Semantic Segmentation with Transformer},
journal = {arXiv preprint arXiv:2210.07124 (2022)},
year = {2022}
}