This repo provides an example code for OCDA-Driving. It implements the AdaptSeg baseline in the paper. This code is heavily borrowed from AdaptSegNet.

• PyTorch (version >= 0.4.1)
• mmcv

• Download the C-Driving Dataset, and put or link it in the root path of this repo.

This script performs distributed training on multiple GPUs.

sh snapshots/vgg16bn_adaptseg/train_dist.sh

This script performs testing. You may change the 'domain' variable in the script to test on different domains.

sh snapshot/vgg16bn_adaptseg/eval_bdd.sh

Note that this repo adopts a different implementation of SyncBN with the original paper, thus the performance may not precisely match that in the paper.

@inproceedings{compounddomainadaptation,
  title={Open Compound Domain Adaptation},
  author={Liu, Ziwei and Miao, Zhongqi and Pan, Xingang and Zhan, Xiaohang and Lin, Dahua and Yu, Stella X. and Gong, Boqing},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2020}
}