Awesome

OSSGAN: Open-Set Semi-Supervised Image Generation

[CVPR 2022] Official pytorch implementation

Paper

Prepare envrioment

To run the code, you need pytorch and some additional packages.

conda env create env.xml
conda activate torch

Quick Start

• Train (-t) and evaluate (-e) the model defined in CONFIG_PATH using GPU 0

CUDA_VISIBLE_DEVICES=0 python3 src/main.py -t -e -c CONFIG_PATH

• Train (-t) and evaluate (-e) the model defined in CONFIG_PATH using GPUs (0, 1, 2, 3) and DataParallel

CUDA_VISIBLE_DEVICES=0,1,2,3 python3 src/main.py -t -e -c CONFIG_PATH

Try python3 src/main.py to see available options.

ImageNet

Prepare dataset

Manual download of the ImageNet dataset (for evaluation and training). Please follow the instructions https://www.tensorflow.org/datasets/catalog/imagenet2012

Put the training and validation set of the ImageNet dataset on ./code/data/ILSVRC2012/{train|valid}.

python3 src/main.py -t -e -l -s -iv -sync_bn -stat_otf -mpc --eval_type valid -c src/configs/ILSVRC2012/BigGAN256.json

Make dataset

python3 src/make_osssimagenet.py --src data/ILSVRC2012 --dst data/OSSSILSVRC2012_50_020_010 --subset_class 50 --ratio 0.2 --usage 0.1

Training

CUDA_VISIBLE_DEVICES=0,...,N python3 src/main.py -t -e -l -sync_bn -stat_otf --eval_type valid -c CONFIG_PATH

Testing

CUDA_VISIBLE_DEVICES=0,...,N python3 src/main.py -e -l -sync_bn -stat_otf --eval_type valid -c CONFIG_PATH --checkpoint_folder CHECKPOINT_FOLDER

Tiny ImageNet

prepare dataset

wget http://cs231n.stanford.edu/tiny-imagenet-200.zip 
unzip tiny-imagenet-200.zip

Put the training and validation set to code/data/TINY_ILSVRC2012/{train|valid}

python3 src/main.py -t -e -l -s -iv -sync_bn -stat_otf -mpc --eval_type valid -c src/configs/TINY_ILSVRC2012/BigGAN-Mod.json

Make datasets

python3 src/make_semi_supervised_dataset.py --src data/TINY_ILSVRC2012 --dst data/OSSSTINY_ILSVRC2012_50_010 --subset_class 50 --ratio 0.1 

Run dataset

python3 src/main.py -t -e -l -sync_bn -stat_otf --eval_type valid -c CONFIG_PATH

License

This repo is built on top of StudioGAN. However, portions of the library are avaiiable under distinct license terms: StyleGAN2, StyleGAN2-ADA, and StyleGAN3 are licensed under NVIDIA source code license, Synchronized batch normalization is licensed under MIT license, HDF5 generator is licensed under MIT license, differentiable SimCLR-style augmentations is licensed under MIT license, and clean-FID is licensed under MIT license.

Bibtex

@misc{katsumata2022ossgan,
      title={OSSGAN: Open-Set Semi-Supervised Image Generation}, 
      author={Kai Katsumata and Duc Minh Vo and Hideki Nakayama},
      year={2022},
      eprint={2204.14249},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}