yutong-gannis / ultra-fast-lane-detection-v2

Ultra Fast Deep Lane Detection With Hybrid Anchor Driven Ordinal Classification (TPAMI 2022)

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

or

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

Tensorrt Deploy

We also provide a python script to do tensorrt inference on videos.

1. Convert to onnx model

   python deploy/pt2onnx.py --config_path configs/culane_res34.py --model_path weights/culane_res34.pth
   

   Or you can download the onnx model using the following script: https://github.com/PINTO0309/PINTO_model_zoo/blob/main/324_Ultra-Fast-Lane-Detection-v2/download.sh. And copy ufldv2_culane_res34_320x1600.onnx to weights/ufldv2_culane_res34_320x1600.onnx

2. Convert to tensorrt model

   Use trtexec to convert engine model

   trtexec --onnx=weights/culane_res34.onnx --saveEngine=weights/culane_res34.engine

3. Do inference

   python deploy/trt_infer.py --config_path  configs/culane_res34.py --engine_path weights/culane_res34.engine --video_path example.mp4
   

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}
}