Home

Awesome

Table of contents
  1. Requirements
  2. Getting Started
  3. Using networks from Python
  4. Preparing datasets
  5. Training
  6. Quality Metrics
  7. Contacts

Efficient Scale-Invariant Generator with Column-Row Entangled Pixel Synthesis

<a href="https://thuanz123.github.io/creps"><img src="https://img.shields.io/badge/WEBSITE-Visit%20project%20page-blue?style=for-the-badge"></a> <a href="https://arxiv.org/abs/2303.14157"><img src="https://img.shields.io/badge/arxiv-2303.14157-red?style=for-the-badge"></a>

Thuan Hoang Nguyen, Thanh Van Le, Anh Tran<br> VinAI Research, Vietnam

Abstract: Any-scale image synthesis offers an efficient and scalable solution to synthesize photo-realistic images at any scale, even going beyond 2K resolution. However, existing GAN-based solutions depend excessively on convolutions and a hierarchical architecture, which introduce inconsistency and the "texture sticking" issue when scaling the output resolution. From another perspective, INR-based generators are scale-equivariant by design, but their huge memory footprint and slow inference hinder these networks from being adopted in large-scale or real-time systems. In this work, we propose Column-Row Entangled Pixel Synthesis (CREPS), a new generative model that is both efficient and scale-equivariant without using any spatial convolutions or coarse-to-fine design. To save memory footprint and make the system scalable, we employ a novel bi-line representation that decomposes layer-wise feature maps into separate "thick" column and row encodings. Experiments on various datasets, including FFHQ, LSUN-Church, MetFaces, and Flickr-Scenery, confirm CREPS' ability to synthesize scale-consistent and alias-free images at any arbitrary resolution with proper training and inference speed.

teaser.png

Details of the model architecture and experimental results can be found in our following paper.

@inproceedings{thuan2023creps,
  title={Efficient Scale-Invariant Generator with Column-Row Entangled Pixel Synthesis},
  author={Thuan Hoang Nguyen, Thanh Van Le, Anh Tran},
  year={2023},
  booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
}

Please CITE our paper whenever our model implementation is used to help produce published results or incorporated into other software.

Additional material

Requirements

The code relies heavily on custom PyTorch extensions that are compiled on the fly using NVCC. On Windows, the compilation requires Microsoft Visual Studio. We recommend installing Visual Studio Community Edition and adding it into PATH using "C:\Program Files (x86)\Microsoft Visual Studio\<VERSION>\Community\VC\Auxiliary\Build\vcvars64.bat".

See Troubleshooting for help on common installation and run-time problems.

Getting started

Pre-trained networks are stored as *.pkl files that can be referenced using local filenames or URLs:

# Generate an image using pre-trained FFHQ model.
python gen_images.py --outdir=out --trunc=1 --seeds=2 \
    --network=creps-ffhq-512x512.pkl --scale=2

# Render a 4x2 grid of interpolations for seeds 0 through 31.
python gen_video.py --output=lerp.mp4 --trunc=1 --seeds=0-31 --grid=4x2 \
    --network=creps-ffhq-512x512.pkl

Outputs from the above commands are placed under out/*.png, controlled by --outdir. Downloaded network pickles are cached under $HOME/.cache/dnnlib, which can be overridden by setting the DNNLIB_CACHE_DIR environment variable. The default PyTorch extension build directory is $HOME/.cache/torch_extensions, which can be overridden by setting TORCH_EXTENSIONS_DIR.

Docker: You can run the above curated image example using Docker as follows:

# Build the creps:latest image
docker build --tag creps .

# Run the gen_images.py script using Docker:
docker run --gpus all -it --rm --user $(id -u):$(id -g) \
    -v `pwd`:/scratch --workdir /scratch -e HOME=/scratch \
    creps \
    python gen_images.py --outdir=out --trunc=1 --seeds=2 \
         --network=creps-ffhq-512x512.pkl

Note: The Docker image requires NVIDIA driver release r470 or later.

The docker run invocation may look daunting, so let's unpack its contents here:

Using networks from Python

You can use pre-trained networks in your own Python code as follows:

with open('ffhq.pkl', 'rb') as f:
    G = pickle.load(f)['G_ema'].cuda()  # torch.nn.Module
z = torch.randn([1, G.z_dim]).cuda()    # latent codes
c = None                                # class labels (not used in this example)
img = G(z, c)                           # NCHW, float32, dynamic range [-1, +1], no truncation

The above code requires torch_utils and dnnlib to be accessible via PYTHONPATH. It does not need source code for the networks themselves — their class definitions are loaded from the pickle via torch_utils.persistence.

The pickle contains three networks. 'G' and 'D' are instantaneous snapshots taken during training, and 'G_ema' represents a moving average of the generator weights over several training steps. The networks are regular instances of torch.nn.Module, with all of their parameters and buffers placed on the CPU at import and gradient computation disabled by default.

The generator consists of two submodules, G.mapping and G.synthesis, that can be executed separately. They also support various additional options:

w = G.mapping(z, c, truncation_psi=0.5, truncation_cutoff=8)
img = G.synthesis(w, noise_mode='const', force_fp32=True)

Please refer to gen_images.py for complete code example.

Preparing datasets

Datasets are stored as uncompressed ZIP archives containing uncompressed PNG files and a metadata file dataset.json for labels. Custom datasets can be created from a folder containing images; see python dataset_tool.py --help for more information. Alternatively, the folder can also be used directly as a dataset, without running it through dataset_tool.py first, but doing so may lead to suboptimal performance.

FFHQ: Download the Flickr-Faces-HQ dataset as 1024x1024 images and create a zip archive using dataset_tool.py:

# Original 1024x1024 resolution.
python dataset_tool.py --source=/tmp/images1024x1024 --dest=~/datasets/ffhq-1024x1024.zip

# Scaled down 256x256 resolution.
python dataset_tool.py --source=/tmp/images1024x1024 --dest=~/datasets/ffhq-256x256.zip \
    --resolution=256x256

See the FFHQ README for information on how to obtain the unaligned FFHQ dataset images. Use the same steps as above to create a ZIP archive for training and validation.

MetFaces: Download the MetFaces dataset and create a ZIP archive:

python dataset_tool.py --source=~/downloads/metfaces/images --dest=~/datasets/metfaces-1024x1024.zip

See the MetFaces README for information on how to obtain the unaligned MetFaces dataset images. Use the same steps as above to create a ZIP archive for training and validation.

AFHQv2: Download the AFHQv2 dataset and create a ZIP archive:

python dataset_tool.py --source=~/downloads/afhqv2 --dest=~/datasets/afhqv2-512x512.zip

Note that the above command creates a single combined dataset using all images of all three classes (cats, dogs, and wild animals), matching the setup used in the CREPS paper. Alternatively, you can also create a separate dataset for each class:

python dataset_tool.py --source=~/downloads/afhqv2/train/cat --dest=~/datasets/afhqv2cat-512x512.zip
python dataset_tool.py --source=~/downloads/afhqv2/train/dog --dest=~/datasets/afhqv2dog-512x512.zip
python dataset_tool.py --source=~/downloads/afhqv2/train/wild --dest=~/datasets/afhqv2wild-512x512.zip

Training

You can train new networks using train.py. For example:

# Train CREPS for FFHQ using 8 GPUs.
python train.py --outdir=~/training-runs --cfg=creps --data=~/datasets/ffhq-512x512.zip \
    --gpus=8 --batch=32 --gamma=8.2 --mirror=1

# Fine-tune CREPS for MetFaces using 1 GPU, starting from the pre-trained FFHQ pickle.
python train.py --outdir=~/training-runs --cfg=creps --data=~/datasets/metfaces-1024x1024.zip \
    --gpus=8 --batch=32 --gamma=6.6 --mirror=1 --kimg=5000 --snap=5 \
    --resume=creps-ffhq-512x512.pkl

Besides CREPS, our code also supports StyleGAN3 and StyleGAN2. Note that the result quality and training time depend heavily on the exact set of options. The most important ones (--gpus, --batch, and --gamma) must be specified explicitly, and they should be selected with care. See python train.py --help for the full list of options and Training configurations for general guidelines & recommendations, along with the expected training speed & memory usage in different scenarios.

The results of each training run are saved to a newly created directory, for example ~/training-runs/00000-creps-afhqv2-512x512-gpus8-batch32-gamma8.2. The training loop exports network pickles (network-snapshot-<KIMG>.pkl) and random image grids (fakes<KIMG>.png) at regular intervals (controlled by --snap). For each exported pickle, it evaluates FID (controlled by --metrics) and logs the result in metric-fid50k_full.jsonl. It also records various statistics in training_stats.jsonl, as well as *.tfevents if TensorBoard is installed.

Quality metrics

By default, train.py automatically computes FID for each network pickle exported during training. We recommend inspecting metric-fid50k_full.jsonl (or TensorBoard) at regular intervals to monitor the training progress. When desired, the automatic computation can be disabled with --metrics=none to speed up the training slightly.

Additional quality metrics can also be computed after the training:

# Previous training run: look up options automatically, save result to JSONL file.
python calc_metrics.py --metrics=fid50k_full \
    --network=~/training-runs/creps-mydataset/network-snapshot-000000.pkl

# Pre-trained network pickle: specify dataset explicitly, print result to stdout.
python calc_metrics.py --metrics=fid50k_full --data=~/datasets/ffhq-1024x1024.zip --mirror=1 \
    --network=creps-ffhq-512x512.pkl --scale=2

The first example looks up the training configuration and performs the same operation as if --metrics=fid50k_full had been specified during training. The second example downloads a pre-trained network pickle, in which case the values of --data and --mirror must be specified explicitly.

Note that the metrics can be quite expensive to compute (up to 1h), and many of them have an additional one-off cost for each new dataset (up to 30min). Also note that the evaluation is done using a different random seed each time, so the results will vary if the same metric is computed multiple times.

Recommended metrics:

Legacy metrics:

References:

  1. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium, Heusel et al. 2017
  2. Demystifying MMD GANs, Bińkowski et al. 2018
  3. Improved Precision and Recall Metric for Assessing Generative Models, Kynkäänniemi et al. 2019
  4. A Style-Based Generator Architecture for Generative Adversarial Networks, Karras et al. 2018
  5. Alias-Free Generative Adversarial Networks, Karras et al. 2021
  6. Improved Techniques for Training GANs, Salimans et al. 2016

Acknowledgments

This repo is built upon StyleGAN3 and we thank the authors for their great works and efforts to release source code. Furthermore, a special "thank you" to CIPS and AnyresGAN, which heavily inspire our work.

Contacts

If you have any questions, please drop an email to v.thuannh5@vinai.io or open an issue in this repository.