Multi domain - Multi modal Image translation

This repository provides Pytorch Implementation of Multidomain-Multimodal Image translation.

Our source code is based on official implementation of StarGANv2

Environment

With basic settings,

conda install -c conda-forge munch
pip install opencv-python
conda install -c anaconda scikit-image

Dataset

For dataset download, please refer to the following link

Seasons / Weather

Before traning / evaluation, split the data into train & validation set,

then save the data in data folder.

To access the train&val dataset that we used, contact : [email protected]

Pre-trained model download

To download our pre-trained model, please refer to the following link

Pre-trained Seasons / Pre-trained Weather

Training

For training, use bash run_train.sh or use the command

or run the command for training models on Seasons dataset

python main_hSRC.py --mode train --num_domains 4 --w_hpf 0 --lambda_reg 1 --lambda_sty 1 --lambda_ds 2 --lambda_cyc 1 \
--lambda_src 0.1 --lambda_nce 0.0 --lambda_dce 1.0 \
--checkpoint_dir ./expr/checkpoints_seasons --eval_dir ./expr/eval_seasons \
--sample_dir ./expr/samples_seasons \
--train_img_dir ./data/alps_proc/train \
--val_img_dir ./data/alps_proc/val --result_dir ./expr/results_seasons \
--use_curriculum --n_patch 128 --use_hard

for Weather dataset

python main_hSRC.py --mode train --num_domains 5 --w_hpf 0 --lambda_reg 1 --lambda_sty 1 --lambda_ds 2 --lambda_cyc 1 \
--lambda_src 0.1 --lambda_nce 0.0 --lambda_dce 0.1 \
--checkpoint_dir ./expr/checkpoints_seasons --eval_dir ./expr/eval_seasons \
--sample_dir ./expr/samples_seasons \
--train_img_dir ./data/alps_proc/train \
--val_img_dir ./data/alps_proc/val --result_dir ./expr/results_seasons \
--use_curriculum --n_patch 128 --use_hard

Evaluation

For evaluation, use bash run_eval.sh or use the command

CUDA_VISIBLE_DEVICES=0 python main_hSRC.py --mode eval --num_domains 4 --w_hpf 0 \
--checkpoint_dir ./expr/checkpoints_seasons --eval_dir ./expr/eval_seasons \
--sample_dir ./expr/samples_seasons \
--train_img_dir ./data/seasons/train \
--val_img_dir ./data/seasons/val --result_dir ./expr/results_seasons --resume_iter 0

or

CUDA_VISIBLE_DEVICES=0 python main_hSRC.py --mode eval --num_domains 5 --w_hpf 0 \
--checkpoint_dir ./expr/checkpoints_weather --eval_dir ./expr/eval_weather \
--sample_dir ./expr/samples_weather \
--train_img_dir ./data/weather/train \
--val_img_dir ./data/weather/train --result_dir ./expr/results_weather --resume_iter 0