/BAPA-Net

Code for BAPA-Net: Boundary Adaptation and Prototype Alignment for Cross-domain Semantic Segmentation

A Pytorch Implementation of BAPA-Net: Boundary Adaptation and Prototype Alignment for Cross-domain Semantic Segmentation (ICCV 2021 oral)

Requirements

pip3 install -r requirements.txt

Usage

  • Download datasets

  • Example of testing a model with domain adaptation with CityScapes as target domain(class_num=16) python3 evaluateUDA.py --model-path *checkpoint.pth* --class-num 16

Checkpoints

We provide the checkpoints at Google Drive.

Citation

@InProceedings{Liu_2021_ICCV,
author = {Liu, Yahao and Deng, Jihong and Gao, Xinchen and Li, Wen and Duan, Lixin},
title = {BAPA-Net: Boundary Adaptation and Prototype Alignment for Cross-domain Semantic Segmentation},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {October},
year = {2021}
}

Acknowledgements

Some codes are adapted from DACS and UDADT. We thank them for their excellent projects.

Contact