Improving Nighttime Driving-Scene Segmentation via Dual Image-adaptive Learnable Filters

[arxiv]

Wenyu Liu, Wentong Li, Jianke Zhu, Miaomiao Cui, Xuansong Xie, Lei Zhang

Requirements

  • python3.7
  • pytorch==1.5.0
  • cuda10.2
  • scikit-image
  • opencv-python

Datasets and Models

Cityscapes: Cityscape NightCity: NightCity ACDC: ACDC Dark-Zurich: Dark-Zurich

Models: Google Drive

Test

Step1: download the [Models](https://drive.google.com/drive/folders/1UDxIAk8v56455XfTB52J2jmgcZEtLAbH?usp=sharing) and put it in the root.
Step2: change the data and model paths in configs/test_config.py
Step3: run "python evaluation_supervised.py" for supervised experiments,  "python evaluation_unsupervised.py" for unsupervised experiments,
Step4: run "python compute_iou.py"

Training

Step1: download the [pre-trained models](https://www.dropbox.com/s/3n1212kxuv82uua/pretrained_models.zip?dl=0) and put it in the root.
Step2: change the data and model paths in configs/train_config.py
Step3: run "python train_unsupervised.py" for unsupervised experiments, run "python train_nightcity.py" for supervised nightcity experiments, run "python train_acdc_night.py" for supervised acdc experiments

Acknowledgments

The code is based on DANNet, PSPNet, Deeplab-v2 and RefineNet.

More works

The image-adaptive filtering techniques used in the detection task can be found in our AAAI2022 paper.

Image-Adaptive YOLO for Object Detection in Adverse Weather Conditions [Link]

Citation

@article{liu2022improving,
  title={Improving Nighttime Driving-Scene Segmentation via Dual Image-adaptive Learnable Filters},
  author={Liu, Wenyu and Li, Wentong and Zhu, Jianke and Cui, Miaomiao and Xie, Xuansong and Zhang, Lei},
  journal={arXiv e-prints},
  pages={arXiv--2207},
  year={2022}
}

GitHub

View Github