hufu6371/gcgan

GcGAN: Geometry-Consistent Generative Adversarial Networks for One-Sided Unsupervised Domain Mapping

Paper

H. Fu, M. Gong, C. Wang, K. Batmanghelich, K. Zhang and D. Tao: Geometry-Consistent Generative Adversarial Networks for One-Sided Unsupervised Domain Mapping. CVPR 2019 (oral)

Huan Fu and Mingming Gong contribute equally to the project and the paper.

Pretrained Models will avaliable soon.

Introduction

The codes have been modified from CycleGAN-Pix2Pix and DistanceGAN, and have been tested successfully on CentOS release 6.9, Cuda 9.1, Tesla V100, Anaconda python3, Pytorch 0.4.1.

The codes are only for research purposes. You may also need to follow the instructions of CycleGAN-Pix2Pix and DistanceGAN.

Usage

1. Clone the respository:

git clone https://github.com/hufu6371/GcGAN.git
cd $GcGAN_ROOT

2. Download the dataset (Cityscapes):

sh ./scripts/download_cyclegan_dataset.sh cityscapes

3. Traning and Test (parsing2city):

sh ./train_gcgan.sh
sh ./test_gcgan.sh

4. Evaluation (parsing2city):

Install pycaffe
Download the pre-trained caffe model following the instructions stated in CycleGAN and Pix2Pix
cd evaluation/parsing2city
python evaluate.py

Pretrained Models

all models

synthetic2real dataset

Tips

1. For city2parsing, we do not apply scale augmentation. (--loadSize 128, --fineSize 128)
2. For parsing2city, we set the hyperparameter "identity" as 0.3. For others, the hyperparameter "identity" is 0.5.
3. For horse2zebra, winter2summer, summer2winter, we do not share parameters for G_{XY} and G_{\tilde{X}\tilde{Y}}. (--model gc_gan_cross)
4. For svhn2mnist, the training scripts are located in the "models/mnist_to_svhn" directory. Please follow the instructions stated in DistanceGAN for training.
5. The evaluation scripts are located in the "$GcGAN_ROOT/evalutation" directory.
6. All the models are trained on 4 GPU cards.

Citation

@inproceedings{FuCVPR19-GcGAN,
  TITLE = {{Geometry-Consistent Generative Adversarial Networks for One-Sided Unsupervised Domain Mapping}},
  AUTHOR = {Fu, Huan and Gong, Mingming and Wang, Chaohui and Batmanghelich, Kayhan and Zhang, Kun and Tao, Dacheng},
  BOOKTITLE = {{IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}},
  YEAR = {2019}
}

Contact

Huan Fu: [email protected]

gcgan's Issues

Pretrained Models

Hi,

I cannot download the pretrained model from the provided google drive link.
I get the following error Sorry, the file you have requested does not exist.
Has the url been moved?

Regards,

Error with multiprocessing?

Hi, thanks for sharing the project. I was interested in playing around with the model and hoping to recreate some of the baseline results.

When running the train_gcgan.sh script I came into a number of errors. I think there might be an error in the data loader/custom data loading process. The error message I'm receiving is the following:

Traceback (most recent call last):
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\spawn.py", line 105, in spawn_main
 File "train.py", line 23, in <module>
   exitcode = _main(fd)
   for i, data in enumerate(dataset):
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\spawn.py", line 114, in _main
 File "C:\Users\manod\Desktop\hand dataset\Geometric_consistent_GAN\data\custom_dataset_data_loader.py", line 44, in __iter__
   prepare(preparation_data)
   for i, data in enumerate(self.dataloader):
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\spawn.py", line 225, in prepare
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\site-packages\torch\utils\data\dataloader.py", line 451, in __iter__
   _fixup_main_from_path(data['init_main_from_path'])
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\spawn.py", line 277, in _fixup_main_from_path
   run_name="__mp_main__")
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\runpy.py", line 263, in run_path
   pkg_name=pkg_name, script_name=fname)
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\runpy.py", line 96, in _run_module_code
   mod_name, mod_spec, pkg_name, script_name)
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\runpy.py", line 85, in _run_code
   exec(code, run_globals)
 File "C:\Users\manod\Desktop\hand dataset\Geometric_consistent_GAN\train.py", line 23, in <module>
   for i, data in enumerate(dataset):
 File "C:\Users\manod\Desktop\hand dataset\Geometric_consistent_GAN\data\custom_dataset_data_loader.py", line 44, in __iter__
   for i, data in enumerate(self.dataloader):
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\site-packages\torch\utils\data\dataloader.py", line 451, in __iter__
   return _DataLoaderIter(self)
   return _DataLoaderIter(self)
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\site-packages\torch\utils\data\dataloader.py", line 239, in __init__
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\site-packages\torch\utils\data\dataloader.py", line 239, in __init__
   w.start()
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\process.py", line 105, in start
   w.start()
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\process.py", line 105, in start
   self._popen = self._Popen(self)
   self._popen = self._Popen(self)
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\context.py", line 223, in _Popen
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\context.py", line 223, in _Popen
   return _default_context.get_context().Process._Popen(process_obj)
   return _default_context.get_context().Process._Popen(process_obj)
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\context.py", line 322, in _Popen
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\context.py", line 322, in _Popen
   return Popen(process_obj)
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\popen_spawn_win32.py", line 65, in __init__
   return Popen(process_obj)
   reduction.dump(process_obj, to_child)
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\popen_spawn_win32.py", line 33, in __init__
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\reduction.py", line 60, in dump
   prep_data = spawn.get_preparation_data(process_obj._name)
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\spawn.py", line 143, in get_preparation_data
   ForkingPickler(file, protocol).dump(obj)
   _check_not_importing_main()
BrokenPipeError: [Errno 32] Broken pipe
 File "C:\Users\manod\Anaconda3\envs\GANtest\lib\multiprocessing\spawn.py", line 136, in _check_not_importing_main
   is not going to be frozen to produce an executable.''')```

Pre-trained model release as promised?

Hi @hufu6371 , thank you so much for this fantastic work.

I finally had the chance to see your team's great presentation from CVPR.

But in the README you mention that you will be releasing the pretrained models soon. I could not find them, so I was wondering what the timeline is.

Thanks a lot!

Typos in gc_cycle_gan_model.py

Hi, thanks a lot for sharing your excellent work.
I find there is a typo in gc_cycle_gan_model.py
rec_B = self.netG_AB(fake_A)
loss_cycle_B = self.criterionCycle(rec_B, self.real_B) * self.opt.lambda_AB
rec_gc_B = self.netG_BA(fake_gc_A)
loss_cycle_B += self.criterionCycle(rec_gc_B, self.real_gc_B) * self.opt.lambda_AB
should be
rec_gc_B = self.netG_AB(fake_gc_A)
I do not know whether it will influence the results.

The link provided to download the pre-trained model

The link provided to download the pre-trained model is no longer available. Could you provide a new one, thank u!