BASeg

The code of paper "BASeg: Boundary Aware Semantic Segmentation for Autonomous Driving"

Introduction

This repository is a official PyTorch implementation for semantic segmentation.

Usage

Requirement:
- Hardware: 4-8 GPUs (better with >=11G GPU memory)
- Software: PyTorch>=1.1.0, Python3, tensorboardX

Clone the repository:

git clone git@github.com:YangParky/BASeg.git

Data preparation
- Download related datasets (ADE20K, Cityscapes, CamVid) and symlink the paths to them as follows (you can alternatively modify the relevant paths specified in folder config):
- To boost the slow speed of the training, you're supposed to prepare the boundary ground truth from here.
- The directory structure is the standard layout for the torchvision
```
/Dataset/
ADE20K/
 Scene-Parsing/
    ADEChallengeData2016/
      images/
      bound/
      annotations/
Cityscapes/
 bound/
 gtFine/
 leftImg8bit/
CamVid/
 bound/
 CamVid_Label/
 CamVid_RGB/
/Model
/Project
/BASeg/
```
Train:
- Download ImageNet pre-trained models and put them under folder model for weight initialization.
- For full traning: ADE20K:
```
sh tools/trainade.sh ade20k baseg101
```
  Cityscapes:
```
sh tools/traincityscapes.sh cityscapes baseg101
```
  CamVid:
```
sh tools/traincamvid.sh camvid baseg101
```
Test:
- Download trained segmentation models and put them under folder specified in config or modify the specified paths.
- For full testing (get listed performance): Validation on ADE20K
```
sh tools/testade.sh ade20k baseg101
```
  Test on Cityscapes
```
sh tools/testcityscapes.sh cityscapes baseg101
```
  Validation on CamVid
```
sh tools/testcamvid.sh camvid baseg101
```
- For boundary evaluation: Evaluation on boundary F1_score
```
python util/f_boundary.py
```
  Evaluation on interior F1_score
```
python util/f_interior.py
```

Citation

If you find the code or trained models useful, please consider citing:

@article{xiao2023baseg,
  title={BASeg: Boundary aware semantic segmentation for autonomous driving},
  author={Xiao, Xiaoyang and Zhao, Yuqian and Zhang, Fan and Luo, Biao and Yu, Lingli and Chen, Baifan and Yang, Chunhua},
  journal={Neural Networks},
  volume={157},
  pages={460--470},
  year={2023},
  publisher={Elsevier}
}

Acknowledgement

The code is from the first author of semseg.

License

This repository is released under MIT License (see LICENSE file for details).

YangParky / BASeg

readme