This is our PyTorch implementation for ToDayGAN. Code was written by Asha Anoosheh (built upon ComboGAN)
If you use this code for your research, please cite:
Night-to-Day Image Translation for Retrieval-based Localization Asha Anoosheh, Torsten Sattler, Radu Timofte, Marc Pollefeys, Luc van Gool In Arxiv, 2018.
pip install torch
pip install torchvision
pip install visdom
pip install dominate
git clone https://github.com/AAnoosheh/ToDayGAN.git
Example running scripts can be found in the scripts
directory.
One of our pretrained models for the Oxford Robotcars dataset is found HERE. Place under ./checkpoints/robotcar_2day and test using the instructions below, with args --name robotcar_2day --dataroot ./datasets/<your_test_dir> --n_domains 2 --which_epoch 150 --loadSize 512
Because of sesitivity to instrinsic camera characteristics, testing should ideally be on the same Oxford dataset photos (and same Grasshopper camera) found conveniently preprocessed and ready-to-use HERE.
If using this pretrained model, <your_test_dir>
should contain two subfolders test0
& test1
, containing Day and Night images to test, respectively (as mine was trained with this ordering). test0
can be empty if you do not care about Day image translated to Night, but just needs to exist to not break the code.
python train.py --name <experiment_name> --dataroot ./datasets/<your_dataset> --n_domains <N> --niter <num_epochs_constant_LR> --niter_decay <num_epochs_decaying_LR>
Checkpoints will be saved by default to ./checkpoints/<experiment_name>/
python train.py --continue_train --which_epoch <checkpoint_number_to_load> --name <experiment_name> --dataroot ./datasets/<your_dataset> --n_domains <N> --niter <num_epochs_constant_LR> --niter_decay <num_epochs_decaying_LR>
python test.py --phase test --serial_test --name <experiment_name> --dataroot ./datasets/<your_dataset> --n_domains <N> --which_epoch <checkpoint_number_to_load>
The test results will be saved to a html file here: ./results/<experiment_name>/<epoch_number>/index.html
.
options/train_options.py
for training-specific flags; see options/test_options.py
for test-specific flags; and see options/base_options.py
for all common flags.--dataroot
) should contain subfolders of the form train*/
and test*/
, and they are loaded in alphabetical order. (Note that a folder named train10 would be loaded before train2, and thus all checkpoints and results would be ordered accordingly.) Test directories should match alphabetical ordering of the training ones.--gpu_ids 0
): set--gpu_ids -1
to use CPU mode; set --gpu_ids 0,1,2
for multi-GPU mode.--display_id
> 0, the results and loss plot will appear on a local graphics web server launched by visdom. To do this, you should have visdom
installed and a server running by the command python -m visdom.server
. The default server URL is http://localhost:8097
. display_id
corresponds to the window ID that is displayed on the visdom
server. The visdom
display functionality is turned on by default. To avoid the extra overhead of communicating with visdom
set --display_id 0
. Secondly, the intermediate results are also saved to ./checkpoints/<experiment_name>/web/index.html
. To avoid this, set the --no_html
flag.--resize_or_crop
option. The default option 'resize_and_crop'
resizes the image such that the largest side becomes opt.loadSize
and then does a random crop of size (opt.fineSize, opt.fineSize)
. Other options are either just resize
or crop
on their own.