Experiments on DA for Semantic Segmentation

Requirements

• Python 3.6+ (recommended, we have not tested the code with previous versions)
• PyTorch 1.6+ (for mixed precision training)
• imgaug and imagecorruptions libraries (refer their installation instructions)
• For simplicity, we provide a environment.yml file extracted from our conda environment. Install using

conda env create -f environment.yml
conda activate daseg

Preparation

Downloading datasets

• Download the Cityscapes dataset
• Download the GTA5 dataset
• Download the SYNTHIA dataset (RAND-CITYSCAPES version)
• Download the SYNTHIA processed labels
• Download the Synscapes dataset

Preparing datasets for training/evaluation

• Create dataset symlinks for GTA5, SYNTHIA, Synscapes, and Cityscapes inside datasets folder:

ln -s /path/to/cityscape ./datasets/cityscape
ln -s /path/to/gta5 ./datasets/gta5-dataset
ln -s /path/to/synthia ./datasets/synthia_cityscape
ln -s /path/to/synscapes ./datasets/synscapes

Pretrained weights

Pretrained weights can be downloaded and copied to the checkpoints folder in either vendorside or clientside (coming soon) folder as required.

• Weights from this paper's results - Google Drive
• Weights from ICCV21 baseline - Google Drive

Evaluation

• Use bash eval.sh within vendorside folder to evaluate any saved model weights.
• Set arguments appropriately in eval.sh file. The important arguments are:
  • CUDA_VISIBLE_DEVICES: GPU ID to be used for training.
  • model: specify deeplab or fcn for model architecture.
  • dataset: specify cityscapes for evaluating on target data.
  • load_model: specify path to model weights to be evaluated, e.g. './checkpoints/dl_allg_gta5.pth'

Training

• Refer to the specific README files in vendorside and clientside folders.

Acknowledgements

We are thankful to FDA, DADA, BDL and AdaptSegNet for releasing their code.

Citation

If you find our work helpful in your research, please cite the following paper:

@InProceedings{pmlr-v162-kundu22a,
  title = 	 {Balancing Discriminability and Transferability for Source-Free Domain Adaptation},
  author =       {Kundu, Jogendra Nath and Kulkarni, Akshay R and Bhambri, Suvaansh and Mehta, Deepesh and Kulkarni, Shreyas Anand and Jampani, Varun and Radhakrishnan, Venkatesh Babu},
  booktitle = 	 {Proceedings of the 39th International Conference on Machine Learning},
  pages = 	 {11710--11728},
  year = 	 {2022},
  volume = 	 {162},
  series = 	 {Proceedings of Machine Learning Research},
  publisher =    {PMLR},
}