Real-time Domain Adaptation in Semantic Segmentation

Andrea Cognolato*, Clauda Cuttano*, Cristina Tortia*

*All authors have contributed equally

Notebooks

1-BiSeNet-training: notebook to run train BiSeNet. Used for comparing epochs and backbones
2-IDDA-Loader: notebook to explore and understand the IDDA dataset. Used for writing the actual dataloader
3-adversarial-training: notebook to train and evaluate BiSeNet with output space adversarial domain adaptation
4-FDA-training: notebook to train and evaluate BiSeNet with FDA. Its results do not appear in the final report
5-FDA-adversarial-training: notebook to train and evaluate BiSeNet with adversarial domain adaptation and FDA
6-FDA-MBT+PSU: notebbok to load 3 FDA+adversarially-trained BiSeNet models and compute their average predictions. Using these predictions we apply a thresolding to generate the pseudolabels needed for self-supervised training.
7-FDA-adversarial-training-PSU: notebook to train and evaluate BiSeNet with adversarial domain adaptation and self-supervised FDA

Name		Name	Last commit message	Last commit date
Latest commit History 17 Commits
dataset		dataset
development-notebooks		development-notebooks
model		model
.gitignore		.gitignore
1-BiSeNet-training.ipynb		1-BiSeNet-training.ipynb
2-IDDA-Loader.ipynb		2-IDDA-Loader.ipynb
3-adversarial-training.ipynb		3-adversarial-training.ipynb
4-FDA-training.ipynb		4-FDA-training.ipynb
5-FDA-adversarial-training.ipynb		5-FDA-adversarial-training.ipynb
6-FDA-MBT+PSU.ipynb		6-FDA-MBT+PSU.ipynb
7-FDA-adversarial-training-PSU.ipynb		7-FDA-adversarial-training-PSU.ipynb
README.md		README.md
SStrain_FDA_adversarial_01.py		SStrain_FDA_adversarial_01.py
classes_info.json		classes_info.json
eval.py		eval.py
loss.py		loss.py
train.py		train.py
train_FDA_adversarial_01.py		train_FDA_adversarial_01.py
utils.py		utils.py