This repository is for ProCA introduced in the following paper

Zhengkai Jiang, Yuxi Li, Ceyuan Yang, Peng Gao, Yabiao Wang, Ying Tai, Chengjie Wang, "Prototypical Contrast Adaptation for Domain Adaptive Semantic Segmentation", ECCV 2022 [arxiv].

• Python 3.6
• Pytorch 1.7.1
• torchvision from master
• yacs
• matplotlib
• GCC >= 4.9
• OpenCV
• CUDA >= 10.1

Step-by-step installation

conda create --name ProCA -y python=3.6
conda activate ProCA

# this installs the right pip and dependencies for the fresh python
conda install -y ipython pip

pip install ninja yacs cython matplotlib tqdm opencv-python imageio mmcv tqdm torchvision==0.8.2 torch==1.7.1

• Download The Cityscapes Dataset
• Download The SYNTHIA-RAND-CITYSCAPES Dataset
• Download The GTAV Dataset

The data folder should be structured as follows:

├── datasets/
│   ├── cityscapes/     
|   |   ├── gtFine/
|   |   ├── leftImg8bit/
│   ├── synthia/
|   |   ├── RAND_CITYSCAPES/
|   |   ├── synthia_label_info.p
│   ├── gtav/
|   |   ├── images/
|   |   ├── labels/
|   |   ├── gtav_label_info.p

Symlink the required dataset

ln -s /path_to_cityscapes_dataset datasets/cityscapes
ln -s /path_to_synthia_dataset datasets/synthia
ln -s /path_to_gtav_dataset datasets/gtav

Generate the label statics file for SYNTHIA and GTAV Datasets by running

python3 datasets/generate_synthia_label_info.py -d datasets/synthia -o datasets/synthia/
python3 datasets/generate_gtav_label_info.py -d datasets/gtav -o datasets/gtav/

python test.py -cfg configs/deeplabv2_r101_ssl_synthia.yaml resume results/model_proca_ssl.pth

python test.py -cfg configs/deeplabv2_r101_ssl.yaml resume results/model_proca_ssl.pth

We provide the training script using 4 Tesla V100 GPUs.

bash run_proca_resnet101_gta5.sh

We also provide Memory-Bank implementation which can be seen in train_memory_bank.

This code is partly based on the open-source implementations from FADA and SDCA.

If you find this code or idea useful, please cite our work:

@inproceedings{jiang2022prototypical,
  title={Prototypical Contrast Adaptation for Domain Adaptive Segmentation},
  author={Jiang, Zhengkai and Li, Yuxi and Yang, Ceyuan and Gao, Peng and Wang, Yabiao and Tai, Ying and Wang, Chengjie},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2022}
}