This is the official code for:

Daiqing Li, Huan Ling, Seung Wook Kim, Karsten Kreis, Adela Barriuso, Sanja Fidler, Antonio Torralba
[Paper] [Bibtex] [Project Page]

• Python 3.7
• Cuda v11.0+
• gcc v7.3.0
• Pytorch 1.9.0+

Our annotation and model are based on BigGAN-512, please download the model from https://tfhub.dev/deepmind/biggan-512/2, store it in ./pretrain folder. Since the original model is trained using TensorFlow, you need to convert the model weights back to Pytorch, following the instructions here https://github.com/ajbrock/BigGAN-PyTorch/tree/master/TFHub. Notice the model is Licensed under Apache-2.0 issued by DeepMind.

We only release our annotations on sampled BigGAN images and images from ImageNet along with its latents used to get the sampled images. For their licenses, please refer to their websites. Notice our dataset release is under the Creative Commons BY-NC 4.0 license by NVIDIA Corporation. Please see License session for details.

1. Download ImageNet from here.
2. Download our annotations annotations.zip and latent codes latents.zip from gdrive. Unzip them into data folder under ./data/.
3. Process images from ImageNet into our dataset format. Run the following script

python prepare_imagenet_images.py --imagenet_dir [path to imagenet dir] --dataset_dir ./data/

4. Prepare images generated from BigGAN. Please download the pretrained weights following this session. And run

python prepare_biggan_images.py --biggan_ckpt ./pretrain/biggan-512.pth --dataset_dir ./data/

After the processing steps, you should have data folder structure like this

data
|
└───annotations
│   |
│   └───biggan512
│   |   │   n01440764
│   |   │   ...
|   └───real-random-list.txt
└───images
│   |
│   └───biggan512
│   |   │   n01440764
│   |   │   ...
|   └───real-random
│       │   n01440764
│       │   ...
└───latents
│   |
│   └───biggan512
│   |   │   n01440764
│   |   │   ...

After the dataset preparation, we now can train BigDatasetGAN to synthesize dataset.

Run the following

python train.py --gan_ckpt ./pretrain/biggan-512.pth \
                --dataset_dir ./data/ \
                --save_dir ./logs/

You can monitor the training progress in tensorboard, as well as the training predictions in logs dir.

By default, the training runs 5k iteration with a batch size of 4, you can adjust it for the best capacity.

After the training is done, we can synthesize ImageNet with pixel-wise labels.

Run the following

python sample_dataset.py --ckpt [path to your pretrained BigDatasetGAN weights] \
                         --save_dir ./dataset_viz/ \
                         --class_idx 225, 200, [you can give it more with ImagetNet class idx] \
                         --samples_per_class 10
                         --z_var 0.9

As an example, here we sample class 225 and 200 with 10 samples each.

For any code dependency related to BigGAN, the license is under the MIT License, see https://github.com/ajbrock/BigGAN-PyTorch/blob/master/LICENSE.

The work BigDatasetGAN code is released under Creative Commons BY-NC 4.0 license, full text at https://creativecommons.org/licenses/by-nc/4.0/legalcode.

The dataset of BigDatasetGAN is released under the Creative Commons BY-NC 4.0 license by NVIDIA Corporation. You can use, redistribute, and adapt the material for non-commercial purposes, as long as you give appropriate credit by citing our paper and indicating any changes that you've made.

Please cite the following paper if you used the code in this repository.

@misc{li2022bigdatasetgan,
      title={BigDatasetGAN: Synthesizing ImageNet with Pixel-wise Annotations}, 
      author={Daiqing Li and Huan Ling and Seung Wook Kim and Karsten Kreis and Adela Barriuso and Sanja Fidler and Antonio Torralba},
      year={2022},
      eprint={2201.04684},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}