Skip to content

VishmiVishara/NAS_cGAN

BranchesTags
 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits

datasets

datasets

 
 

exp/nas_cgan

exp/nas_cgan

 
 

template_lib @ c7addbe

template_lib @ c7addbe

 
 

.gitignore

.gitignore

 
 

.gitmodules

.gitmodules

 
 

README.md

README.md

 
 

requirements.txt

requirements.txt

 
 

Repository files navigation

Searching towards Class-Aware Generators for Conditional Generative Adversarial Networks paper

Updates

  • Support DistributedDataParallel (2020-09-17)
  • Evaluation code
  • Search code (coming!)

Environment

  • CUDA 10.0; cudnn-10.0-v7.6.5; Python == 3.6
  • PyTorch == 1.4.0, torchvision == 0.5.0
pip install torch==1.4.0+cu100 torchvision==0.5.0+cu100 -f https://download.pytorch.org/whl/torch_stable.html
  • detectron2
python -m pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu100/index.html
  • TensorFlow == 1.13.1
pip install tensorflow-gpu==1.13.1
  • Additional used packages
pip install -r requirements.txt

Clone git repository

git clone --recursive https://github.com/PeterouZh/NAS_cGAN.git
cd NAS_cGAN

Prepare FID statistic file

Pre-calculate FID statistic files of CIFAR10 using PyTorch and TensorFlow models, respectively. Those files will be saved to ./datasets/nas_cgan/pytorch_fid_stat/fid_stats_pytorch_cifar10_train_32.npz and ./datasets/nas_cgan/tf_fid_stat/fid_stats_tf_cifar10_train_32.npz.

Or you can download our pre-calculated statistic files onedrive, which should be put into the directories mentioned above.

export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:/usr/local/cudnn-10.0-v7.6.5.32/lib64:$LD_LIBRARY_PATH
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH=.:./exp
python 	exp/nas_cgan/scripts/train_net.py \
  --config exp/nas_cgan/configs/calculate_fid_stat_CIFAR10.yaml \
  --command calculate_fid_stat_CIFAR10 \
  --outdir results/calculate_fid_stat_CIFAR10

Evaluate the models reported in the paper

Download the trained models onedrive, and put them into ./datasets/nas_cgan/models.

FID and IS

Eval NAS-cGAN on CIFAR10.

export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:/usr/local/cudnn-10.0-v7.6.5.32/lib64:$LD_LIBRARY_PATH
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH=.:./exp
python 	exp/nas_cgan/scripts/train_net.py \
  --config exp/nas_cgan/configs/eval_NAS_cGAN_CIFAR10.yaml \
  --command eval_NAS_cGAN_CIFAR10 \
  --outdir results/eval_NAS_cGAN_CIFAR10

Eval NAS-caGAN on CIFAR10.

export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:/usr/local/cudnn-10.0-v7.6.5.32/lib64:$LD_LIBRARY_PATH
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH=.:./exp
python 	exp/nas_cgan/scripts/train_net.py \
  --config exp/nas_cgan/configs/eval_NAS_caGAN_CIFAR10.yaml \
  --command eval_NAS_caGAN_CIFAR10 \
  --outdir results/eval_NAS_caGAN_CIFAR10

intra FIDs

Pre-calculate intra FID statistic of each class on CIFAR10. Or you can use our pre-calculated files onedrive, and put these files in ./datasets/nas_cgan/tf_fid_stat/cifar10_train_per_class_32.

export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:/usr/local/cudnn-10.0-v7.6.5.32/lib64:$LD_LIBRARY_PATH
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH=.:./exp
python 	exp/nas_cgan/scripts/train_net.py \
  --config exp/nas_cgan/configs/calculate_fid_stat_per_class_CIFAR10.yaml \
  --command calculate_fid_stat_per_class_CIFAR10 \
  --outdir results/calculate_fid_stat_per_class_CIFAR10

Eval intra FIDs for NAS-cGAN.

export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:/usr/local/cudnn-10.0-v7.6.5.32/lib64:$LD_LIBRARY_PATH
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH=.:./exp
python 	exp/nas_cgan/scripts/train_net.py \
  --config exp/nas_cgan/configs/eval_intra_FID_NAS_cGAN_CIFAR10.yaml \
  --command eval_intra_FID_NAS_cGAN_CIFAR10 \
  --outdir results/eval_intra_FID_NAS_cGAN_CIFAR10

Eval intra FIDs for NAS-caGAN.

export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:/usr/local/cudnn-10.0-v7.6.5.32/lib64:$LD_LIBRARY_PATH
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH=.:./exp
python 	exp/nas_cgan/scripts/train_net.py \
  --config exp/nas_cgan/configs/eval_intra_FID_NAS_caGAN_CIFAR10.yaml \
  --command eval_intra_FID_NAS_caGAN_CIFAR10 \
  --outdir results/eval_intra_FID_NAS_caGAN_CIFAR10

The calibrated results

Prepare intra FID statistic files following the steps above (put these files in ./datasets/nas_cgan/tf_fid_stat/cifar10_train_per_class_32). Download FID logits files onedrive produced by the calibrated models, and put the files into ./datasets/nas_cgan/fid_logits_of_calibrated_model/.

export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:/usr/local/cudnn-10.0-v7.6.5.32/lib64:$LD_LIBRARY_PATH
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH=.:./exp
python 	exp/nas_cgan/scripts/train_net.py \
  --config exp/nas_cgan/configs/eval_calibrated_NAS_cGAN.yaml \
  --command eval_calibrated_NAS_cGAN \
  --outdir results/eval_calibrated_NAS_cGAN
export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:/usr/local/cudnn-10.0-v7.6.5.32/lib64:$LD_LIBRARY_PATH
export CUDA_VISIBLE_DEVICES=0
export PYTHONPATH=.:./exp
python 	exp/nas_cgan/scripts/train_net.py \
  --config exp/nas_cgan/configs/eval_calibrated_NAS_caGAN.yaml \
  --command eval_calibrated_NAS_caGAN \
  --outdir results/eval_calibrated_NAS_caGAN

Acknowledgement

  1. https://github.com/facebookresearch/detectron2
  2. https://github.com/TDeVries/enas_pytorch
  3. https://github.com/TAMU-VITA/AutoGAN
  4. https://github.com/ajbrock/BigGAN-PyTorch
  5. https://github.com/rosinality/stylegan2-pytorch
  6. https://github.com/openai/improved-gan/tree/master/inception_score
  7. https://github.com/bioinf-jku/TTUR

About

cGAN with NAS.

Resources

Readme
Activity

Stars

0 stars

Watchers

1 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%