Official code repository for the paper Semantic-Preserving Unsupervised Domain Translation.
Dependencies are provided in requirement.txt. A Dockerfile is provided for reproducing the same environment used to run the experiments.
The models are in src/models. Every model has 3 files:
__init__.py: Defines the specific parameters of the models
model.py Defines the architecture of the model
train.py Defines the training algorithm of the model
A model can be run by invoking main.py. The syntax is as follows
python src/main.py [GENERAL PARAMETERS] [MODEL] [SPECIFIC MODEL PARAMETERS]
As example, we provide the command for reproducing our results on MNIST to SVHN
MNIST
python src/main.py --exp-name cluster --cuda --run-id mnist vrinv --dataset1 imnist --lr 1e-3 --h-dim 256 --ld 1
MNIST-to-SVHN
python src/main.py --exp-name vmt-cluster --cuda --run-id mnist-svhn vmt_cluster --dataset1 mnist --dataset2 svhn --cluster-model-path ./experiments/vrinv/cluster_mnist-None --cluster-model vrinv --dw 0.01 --svw 1 --tvw 0.06 --tcw 0.06 --smw 1 --tmw 0.06
SVHN
python src/main.py --exp-name classifier --cuda --run-id svhn --train-batch-size 128 --valid-split 0.2 classifier --dataset svhn_extra
MNIST-to-SVHN
python src/main.py --run-id mnist-svhn --exp-name UDT --train-batch-size 64 --test-batch-size 50 --cuda udt --eval-model-path ./experiments/classifier/classifier_svhn-None/ --dataset1 mnist --dataset2 svhn --semantic-model-path ./experiments/vmt_cluster/vmt-cluster_mnist-svhn-None --gsxy 0.5
We use tensorboard for saving the artefacts. It is possible to view the results by simply invoking tensorboard in the folder where the results were saved
tensorboard --logdir .