GAL-DeepLabv3Plus

Introduction

This is the official PyTorch implementation of Graph Attention Layer Evolves Semantic Segmentation for Road Pothole Detection: A Benchmark and Algorithms, published on IEEE T-IP in 2021.

In this repository, we provide the training and testing setups on the pothole dataset (paper). We have tested our code in Python 3.8.10, CUDA 11.1, and PyTorch 1.10.1.

Setup

Please setup the pothole dataset and the pretrained weight according to the following folder structure:

GAL-DeepLabv3plus
 |-- data
 |-- datasets
 |  |-- pothole
 |-- models
 |-- options
 |-- runs
 |  |-- tdisp_gal
 ...

The pothole dataset datasets/pothole can be downloaded from here, and the pretrained weight runs/tdisp_gal for our GAL-DeepLabv3+ can be downloaded from here.

Usage

Testing on the Pothole Dataset

For testing, please first setup the runs/tdisp_gal and the datasets/pothole folders as mentioned above. Then, run the following script:

bash ./scripts/test_gal.sh

to test GAL-DeepLabv3+ with the transformed disparity images. The prediction results are stored in testresults.

Training on the Pothole Dataset

For training, please first setup the datasets/pothole folder as mentioned above. Then, run the following script:

bash ./scripts/train_gal.sh

to train GAL-DeepLabv3+ with the transformed disparity images. The weights and the tensorboard record containing the loss curves as well as the performance on the validation set will be saved in runs.

Citation

If you use this code for your research, please cite our paper:

@article{fan2021graph,
  title     = {Graph Attention Layer Evolves Semantic Segmentation for Road Pothole Detection: A Benchmark and Algorithms},
  author    = {Fan, Rui and Wang, Hengli and Wang, Yuan and Liu, Ming and Pitas, Ioannis},
  journal   = {IEEE Transactions on Image Processing},
  volume    = {30},
  number    = {},
  pages     = {8144-8154},
  year      = {2021},
  publisher = {IEEE},
  doi       = {10.1109/TIP.2021.3112316}
}

If you use the pothole dataset for your research, please cite our papers:

@article{fan2019pothole,
  title={Pothole detection based on disparity transformation and road surface modeling},
  author={Fan, Rui and Ozgunalp, Umar and Hosking, Brett and Liu, Ming and Pitas, Ioannis},
  journal={IEEE Transactions on Image Processing},
  volume={29},
  pages={897--908},
  year={2019},
  publisher={IEEE}
}
@article{fan2019road,
  title={Road damage detection based on unsupervised disparity map segmentation},
  author={Fan, Rui and Liu, Ming},
  journal={IEEE Transactions on Intelligent Transportation Systems},
  volume={21},
  number={11},
  pages={4906--4911},
  year={2019},
  publisher={IEEE}
}
@article{fan2018road,
  title={Road surface 3D reconstruction based on dense subpixel disparity map estimation},
  author={Fan, Rui and Ai, Xiao and Dahnoun, Naim},
  journal={IEEE Transactions on Image Processing},
  volume={27},
  number={6},
  pages={3025--3035},
  year={2018},
  publisher={IEEE}
}

Acknowledgement

Our code is inspired by pytorch-CycleGAN-and-pix2pix, pytorch_segmentation, pytorch-deeplab-xception , and RTFNet.