Ultra-Fast-Lane-Detection-V2-plus

优化内容

减低了模型参数; 相对收敛加快; 推理速度更快、精度更高。直接使用官方的train.py就可以训练。更多内容看doc里面的东西。ufld-v3在路上，在等一个长假完成。这是个临时版本，后面我对代码做了i修改。春节时间上传。 Ultra-Fast-Lane-Detection-V3代码已开源，在这个基础上，对车道线做分类。 Tips：发现模型特别大怎么办？把用于保存模型的字典中'optimizer'去掉。

新闻

我在做一个"Fantastic-yolov5"的项目,用它做了很多事，比如分割，检测车道线，分类，ocr识别车牌，多任务。目前做了50%，我在等一个长的假期去把它完成。期待吧。

Ultra-Fast-Lane-Detection-V2

PyTorch implementation of the paper "Ultra Fast Deep Lane Detection with Hybrid Anchor Driven Ordinal Classification".

Demo

Install

Please see INSTALL.md

Get started

Please modify the data_root in any configs you would like to run. We will use configs/culane_res18.py as an example.

To train the model, you can run:

python train.py configs/culane_res18.py --log_path /path/to/your/work/dir

python train.py configs/culane_res18.py --log_path ./work

python -m torch.distributed.launch --nproc_per_node=8 train.py configs/culane_res18.py --log_path /path/to/your/work/dir

It should be noted that if you use different number of GPUs, the learning rate should be adjusted accordingly. The configs' learning rates correspond to 8-GPU training on CULane and CurveLanes datasets. If you want to train on CULane or CurveLanes with single GPU, please decrease the learning rate by a factor of 1/8. On the Tusimple, the learning rate corresponds to single GPU training.

Trained models

We provide trained models on CULane, Tusimple, and CurveLanes.

Dataset	Backbone	F1	Link
CULane	ResNet18	75.0	Google/Baidu
CULane	ResNet34	76.0	Google/Baidu
Tusimple	ResNet18	96.11	Google/Baidu
Tusimple	ResNet34	96.24	Google/Baidu
CurveLanes	ResNet18	80.42	Google/Baidu
CurveLanes	ResNet34	81.34	Google/Baidu

For evaluation, run

mkdir tmp

python test.py configs/culane_res18.py --test_model /path/to/your/model.pth --test_work_dir ./tmp

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

mkdir tmp

python -m torch.distributed.launch --nproc_per_node=8 test.py configs/culane_res18.py --test_model /path/to/your/model.pth --test_work_dir ./tmp

Visualization

We provide a script to visualize the detection results. Run the following commands to visualize on the testing set of CULane.

python demo.py configs/culane_res18.py --test_model /path/to/your/culane_res18.pth

Citation

@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{qin2022ultrav2,
  author={Qin, Zequn and Zhang, Pengyi and Li, Xi},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
  title={Ultra Fast Deep Lane Detection With Hybrid Anchor Driven Ordinal Classification}, 
  year={2022},
  volume={},
  number={},
  pages={1-14},
  doi={10.1109/TPAMI.2022.3182097}
}

scott-mao / Ultra-Fast-Lane-Detection-v2-plus