StyleGAN3-Fun
_{Let's have fun with StyleGAN2/ADA/3!}

SOTA GANs are hard to train and to explore, and StyleGAN2/ADA/3 are no different. The point of this repository is to allow the user to both easily train and explore the trained models without unnecessary headaches.

As before, we will build upon the official repository, which has the advantage of being backwards-compatible. As such, we can use our previously-trained models from StyleGAN2 and StyleGAN2-ADA. Please get acquainted with the official repository and its codebase, as we will be building upon it and as such, increase its capabilities (but hopefully not its complexity!).

Additions

This repository adds/has the following changes (not yet the complete list):

Dataset Setup (dataset_tool.py)
- RGBA support, so revert saving images to .png (Issue #156 by @1378dm). Training can use RGBA and images can be generated.
  - TODO: ~~Check that training code is correct for normalizing the alpha channel~~, as well as making the interpolation code work with this new format (look into moviepy.editor.VideoClip)
  - For now, interpolation videos will only be saved in RGB format, e.g., discarding the alpha channel.
- --center-crop-tall: add vertical black bars to the sides of each image in the dataset (rectangular images, with height > width), and you wish to train a square model, in the same vein as the horizontal bars added when using --center-crop-wide (where width > height).
  - This is useful when you don't want to lose information from the left and right side of the image by only using the center crop (ibidem for --center-crop-wide, but for the top and bottom of the image)
  - Note that each image doesn't have to be of the same size, and the added bars will only ensure you get a square image, which will then be resized to the model's desired resolution (set by --resolution).
- Grayscale images in the dataset are converted to RGB
  - If you want to turn this off, remove the respective line in dataset_tool.py, e.g., if your dataset is made of images in a folder, then the function to be used is open_image_folder in dataset_tool.py, and the line to be removed is img = img.convert('RGB') in the iterate_images inner function.
- The dataset can be forced to be of a specific number of channels, that is, grayscale, RGB or RGBA.
  - To use this, set --force-channels=1 for grayscale, --force-channels=3 for RGB, and --force-channels=4 for RGBA.
- If the dataset tool encounters an error, print it along the offending image, but continue with the rest of the dataset (PR #39 from Andreas Jansson).
- For conditional models, we can use the subdirectories as the classes by adding --subfolders-as-labels. This will generate the dataset.json file automatically as done by @pbaylies here
  - Additionally, in the --source folder, we will save a class_labels.txt file, to further know which classes correspond to each subdirectory.
Training
- Add --cfg=stylegan2-ext, which uses @aydao's extended modifications for handling large and diverse datasets.
  - A good explanation is found in Gwern's blog here
  - If you wish to fine-tune from @aydao's Anime model, use --cfg=stylegan2-ext --resume=anime512 when running train.py
  - Note: This is an extremely experimental configuration! The .pkl files will be ~1.1Gb each and training will slow down significantly. Use at your own risk!
- --blur-percent: Blur both real and generated images before passing them to the Discriminator.
  - The blur (blur_init_sigma=10.0) will completely fade after the selected percentage of the training is completed (using a linear ramp).
  - Another experimental feature, should help with datasets that have a lot of variation, and you wish the model to slowly learn to generate the objects and then its details.
- --mirrory: Added vertical mirroring for doubling the dataset size (quadrupling if --mirror is used; make sure your dataset has either or both of these symmetries in order for it to make sense to use them)
- --gamma: If no R1 regularization is provided, the heuristic formula from StyleGAN will be used.
  - Specifically, we will set gamma=0.0002 * resolution ** 2 / batch_size
- --aug: TODO: add Deceive-D/APA as an option.
- --augpipe: Now available to use is StyleGAN2-ADA's full list of augpipe, i.e., individual augmentations (blit, geom, color, filter, noise, cutout) or their combinations (bg, bgc, bgcf, bgcfn, bgcfnc).
- --img-snap: Set when to save snapshot images, so now it's independent of when the model is saved (e.g., save image snapshots more often to know how the model is training without saving the model itself, to save space).
- --snap-res: The resolution of the snapshots, depending on how many images you wish to see per snapshot. Available resolutions: 1080p, 4k, and 8k.
- --resume-kimg: Starting number of kimg, useful when continuing training a previous run
- --outdir: Automatically set as training-runs, so no need to set beforehand (in general this is true throughout the repository)
- --metrics: Now set by default to None, so there's no need to worry about this one
- --freezeD: Renamed --freezed for better readability
- --freezeM: Freeze the first layers of the Mapping Network Gm (G.mapping)
- --freezeE: Freeze the embedding layer of the Generator (for class-conditional models)
- --freezeG: TODO: Freeze the first layers of the Synthesis Network (G.synthesis; less cost to transfer learn, focus on high layers?)
- --resume: All available pre-trained models from NVIDIA (and more) can be used with a simple dictionary, depending on the --cfg used. For example, if you wish to use StyleGAN3's config-r, then set --cfg=stylegan3-r. In addition, if you wish to transfer learn from FFHQU at 1024 resolution, set --resume=ffhqu1024.
  - The full list of currently available models to transfer learn from (or synthesize new images with) is the following (TODO: add small description of each model, so the user can better know which to use for their particular use-case; proper citation to original authors as well):
    StyleGAN2 models
    1. Majority, if not all, are config-f: set --cfg=stylegan2
      
      ffhq256
      
      ffhqu256
      
      ffhq512
      
      ffhq1024
      
      ffhqu1024
      
      celebahq256
      
      lsundog256
      
      afhqcat512
      
      afhqdog512
      
      afhqwild512
      
      afhq512
      
      brecahad512
      
      cifar10 (conditional, 10 classes)
      
      metfaces1024
      
      metfacesu1024
      
      lsuncar512 (config-f)
      
      lsuncat256 (config-f)
      
      lsunchurch256 (config-f)
      
      lsunhorse256 (config-f)
      
      minecraft1024 (thanks to @jeffheaton)
      
      imagenet512 (thanks to @shawwn)
      
      wikiart1024-C (conditional, 167 classes; thanks to @pbaylies)
      
      wikiart1024-U (thanks to @pbaylies)
      
      maps1024 (thanks to @tjukanov)
      
      fursona512 (thanks to @arfafax)
      
      mlpony512 (thanks to @arfafax)
      
      lhq1024 (thanks to @justinpinkney)
      
      afhqcat256 (Deceive-D/APA models)
      
      anime256 (Deceive-D/APA models)
      
      cub256 (Deceive-D/APA models)
      
      sddogs1024 (Self-Distilled StyleGAN models)
      
      sdelephant512 (Self-Distilled StyleGAN models)
      
      sdhorses512 (Self-Distilled StyleGAN models)
      
      sdbicycles256 (Self-Distilled StyleGAN models)
      
      sdlions512 (Self-Distilled StyleGAN models)
      
      sdgiraffes512 (Self-Distilled StyleGAN models)
      
      sdparrots512 (Self-Distilled StyleGAN models)
    2. Extended StyleGAN2 config from @aydao: set --cfg=stylegan2-ext
      
      anime512 (thanks to @aydao; writeup by @gwern: https://gwern.net/Faces#extended-stylegan2-danbooru2019-aydao)
    StyleGAN3 models
    1. config-t: set --cfg=stylegan3-t
      
      afhq512
      
      ffhqu256
      
      ffhq1024
      
      ffhqu1024
      
      metfaces1024
      
      metfacesu1024
      
      landscapes256 (thanks to @justinpinkney)
      
      wikiart1024 (thanks to @justinpinkney)
      
      mechfuture256 (thanks to @edstoica; 29 kimg tick)
      
      vivflowers256 (thanks to @edstoica; 68 kimg tick)
      
      alienglass256 (thanks to @edstoica; 38 kimg tick)
      
      scificity256 (thanks to @edstoica; 210 kimg tick)
      
      scifiship256 (thanks to @edstoica; 168 kimg tick)
    2. config-r: set --cfg=stylegan3-r
      
      afhq512
      
      ffhq1024
      
      ffhqu1024
      
      ffhqu256
      
      metfaces1024
      
      metfacesu1024
  - The main sources of these pretrained models are both the official NVIDIA repository, as well as other community repositories, such as Justin Pinkney 's Awesome Pretrained StyleGAN2 and Awesome Pretrained StyleGAN3, Deceive-D/APA, Self-Distilled StyleGAN/Internet Photos, and edstoica 's Wombo Dream -based models. Others can be found around the net and are properly credited in this repository, so long as they can be easily downloaded with dnnlib.util.open_url.
Interpolation videos
- Random interpolation
  - Generate images/interpolations with the internal representations of the model
    - Usage: Add --layer=<layer_name> to specify which layer to use for interpolation.
    - If you don't know the names of the layers available for your model, add the flag --available-layers and the layers will be printed to the console, along their names, number of channels, and sizes.
    - Use one of --grayscale or --rgb to specify whether to save the images as grayscale or RGB during the interpolation.
    - For --rgb, three consecutive channels (starting at --starting-channel=0) will be used to create the RGB image. For --grayscale, only the first channel will be used.
- Style-mixing
- Sightseeding (jumpiness has been fixed)
- Circular interpolation
- Visual-reactive interpolation (Beta)
- Audiovisual-reactive interpolation (TODO)
- TODO: Give support to RGBA models!
Projection into the latent space
- Project into $\mathcal{W}+$
- Additional losses to use for better projection (e.g., using VGG16 or CLIP)
Discriminator Synthesis (official code)
- Generate a static image (python discriminator_synthesis.py dream --help) or a video with a feedback loop (python discriminator_synthesis.py dream-zoom --help, python discriminator_synthesis.py channel-zoom --help, or python discriminator_synthesis.py interp --help)
- Start from a random image (random for noise or perlin for 2D fractal Perlin noise, using Mathieu Duchesneau's implementation) or from an existing one
Expansion on GUI/visualizer.py
- Added the rest of the affine transformations
- Added widget for class-conditional models (TODO: mix classes with continuous values for cls!)
General model and code additions
- Multi-modal truncation trick: find the different clusters in your model and use the closest one to your dlatent, in order to increase the fidelity
  - Usage: Run python multimodal_truncation.py get-centroids --network=<path_to_model> to use default values; for extra options, run python multimodal_truncation.py get-centroids --help
- StyleGAN3: anchor the latent space for easier to follow interpolations (thanks to Rivers Have Wings and nshepperd).
- Use CPU instead of GPU if desired (not recommended, but perfectly fine for generating images, whenever the custom CUDA kernels fail to compile).
- Add missing dependencies and channels so that the conda environment is correctly setup in Windows (PR's #111/#125 and #80 /#143 from the base, respectively)
- Use StyleGAN-NADA models with any part of the code (Issue #9)
  - The StyleGAN-NADA models must first be converted via Vadim Epstein 's conversion code found here.
- Add PR #173 for adding the last remaining unknown kwarg for using StyleGAN2 models using TF 1.15.
TODO list (this is a long one with more to come, so any help is appreciated):
- Add --device={cuda, ref} option to pass to each of the custom operations in order to (theoretically) be able to use AMD GPUs, as explained in @l4rz's post here
- Define a custom Generator in legacy.py to modify the output size
  - Related: the multi-latent, i.e., the one from @eps696
- Add Top-K training as done here and here
- Add panorama/SinGAN/feature interpolation from StyleGAN of All Trades
- PTI for better inversion
- Better sampling
- Add cross-model interpolation
- Blend different models (average checkpoints, copy weights, create initial network), as in @aydao's StyleGAN2-Surgery
- Add multi-crop for the dataset creation, as used in Earth View.
- Make it easy to download pretrained models from Drive, otherwise a lot of models can't be used with dnnlib.util.open_url (e.g., StyleGAN-Human models)
- Finish documentation for better user experience, add videos/images, code samples, visuals...
- Add Ensembling Off-the-shelf Models for GAN Training and Any-resolution Training for High-resolution Image Synthesis

Notebooks (Coming soon!)

Alias-Free Generative Adversarial Networks (StyleGAN3)
_{Official PyTorch implementation of the NeurIPS 2021 paper}

Alias-Free Generative Adversarial Networks
Tero Karras, Miika Aittala, Samuli Laine, Erik Härkönen, Janne Hellsten, Jaakko Lehtinen, Timo Aila
https://nvlabs.github.io/stylegan3

Abstract: We observe that despite their hierarchical convolutional nature, the synthesis process of typical generative adversarial networks depends on absolute pixel coordinates in an unhealthy manner. This manifests itself as, e.g., detail appearing to be glued to image coordinates instead of the surfaces of depicted objects. We trace the root cause to careless signal processing that causes aliasing in the generator network. Interpreting all signals in the network as continuous, we derive generally applicable, small architectural changes that guarantee that unwanted information cannot leak into the hierarchical synthesis process. The resulting networks match the FID of StyleGAN2 but differ dramatically in their internal representations, and they are fully equivariant to translation and rotation even at subpixel scales. Our results pave the way for generative models better suited for video and animation.

For business inquiries, please visit our website and submit the form: NVIDIA Research Licensing

Release notes

This repository is an updated version of stylegan2-ada-pytorch, with several new features:

Alias-free generator architecture and training configurations (stylegan3-t, stylegan3-r).
Tools for interactive visualization (visualizer.py), spectral analysis (avg_spectra.py), and video generation (gen_video.py).
Equivariance metrics (eqt50k_int, eqt50k_frac, eqr50k).
General improvements: reduced memory usage, slightly faster training, bug fixes.

Compatibility:

Compatible with old network pickles created using stylegan2-ada and stylegan2-ada-pytorch. (Note: running old StyleGAN2 models on StyleGAN3 code will produce the same results as running them on stylegan2-ada/stylegan2-ada-pytorch. To benefit from the StyleGAN3 architecture, you need to retrain.)
Supports old StyleGAN2 training configurations, including ADA and transfer learning. See Training configurations for details.
Improved compatibility with Ampere GPUs and newer versions of PyTorch, CuDNN, etc.

Synthetic image detection

While new generator approaches enable new media synthesis capabilities, they may also present a new challenge for AI forensics algorithms for detection and attribution of synthetic media. In collaboration with digital forensic researchers participating in DARPA's SemaFor program, we curated a synthetic image dataset that allowed the researchers to test and validate the performance of their image detectors in advance of the public release. Please see here for more details.

Additional material

Result videos
Curated example images
StyleGAN3 pre-trained models for config T (translation equiv.) and config R (translation and rotation equiv.)

_{Access individual networks via https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/<MODEL>, where <MODEL> is one of:}
_{stylegan3-t-ffhq-1024x1024.pkl, stylegan3-t-ffhqu-1024x1024.pkl, stylegan3-t-ffhqu-256x256.pkl}
_{stylegan3-r-ffhq-1024x1024.pkl, stylegan3-r-ffhqu-1024x1024.pkl, stylegan3-r-ffhqu-256x256.pkl}
_{stylegan3-t-metfaces-1024x1024.pkl, stylegan3-t-metfacesu-1024x1024.pkl}
_{stylegan3-r-metfaces-1024x1024.pkl, stylegan3-r-metfacesu-1024x1024.pkl}
_{stylegan3-t-afhqv2-512x512.pkl}
_{stylegan3-r-afhqv2-512x512.pkl}
StyleGAN2 pre-trained models compatible with this codebase

_{Access individual networks via https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan2/versions/1/files/<MODEL>, where <MODEL> is one of:}
_{stylegan2-ffhq-1024x1024.pkl, stylegan2-ffhq-512x512.pkl, stylegan2-ffhq-256x256.pkl}
_{stylegan2-ffhqu-1024x1024.pkl, stylegan2-ffhqu-256x256.pkl}
_{stylegan2-metfaces-1024x1024.pkl, stylegan2-metfacesu-1024x1024.pkl}
_{stylegan2-afhqv2-512x512.pkl}
_{stylegan2-afhqcat-512x512.pkl, stylegan2-afhqdog-512x512.pkl, stylegan2-afhqwild-512x512.pkl}
_{stylegan2-brecahad-512x512.pkl, stylegan2-cifar10-32x32.pkl}
_{stylegan2-celebahq-256x256.pkl, stylegan2-lsundog-256x256.pkl}

Requirements

Linux and Windows are supported, but we recommend Linux for performance and compatibility reasons.
1–8 high-end NVIDIA GPUs with at least 12 GB of memory. We have done all testing and development using Tesla V100 and A100 GPUs.
64-bit Python 3.8 and PyTorch 1.9.0 (or later). See https://pytorch.org for PyTorch install instructions.
CUDA toolkit 11.1 or later. (Why is a separate CUDA toolkit installation required? See Troubleshooting).
GCC 7 or later (Linux) or Visual Studio (Windows) compilers. Recommended GCC version depends on CUDA version, see for example CUDA 11.4 system requirements.
Python libraries: see environment.yml for exact library dependencies. You can use the following commands with Miniconda3 to create and activate your StyleGAN3 Python environment:
- conda env create -f environment.yml
- conda activate stylegan3
Docker users:
- Ensure you have correctly installed the NVIDIA container runtime.
- Use the provided Dockerfile to build an image with the required library dependencies.

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 AFHQv2 model ("Ours" in Figure 1, left).
python gen_images.py --outdir=out --trunc=1 --seeds=2 \
    --network=https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-r-afhqv2-512x512.pkl

# 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=https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-r-afhqv2-512x512.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 stylegan3:latest image
docker build --tag stylegan3 .

# 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 \
    stylegan3 \
    python gen_images.py --outdir=out --trunc=1 --seeds=2 \
         --network=https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-r-afhqv2-512x512.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:

--gpus all -it --rm --user $(id -u):$(id -g): with all GPUs enabled, run an interactive session with current user's UID/GID to avoid Docker writing files as root.
-v `pwd`:/scratch --workdir /scratch: mount current running dir (e.g., the top of this git repo on your host machine) to /scratch in the container and use that as the current working dir.
-e HOME=/scratch: let PyTorch and StyleGAN3 code know where to cache temporary files such as pre-trained models and custom PyTorch extension build results. Note: if you want more fine-grained control, you can instead set TORCH_EXTENSIONS_DIR (for custom extensions build dir) and DNNLIB_CACHE_DIR (for pre-trained model download cache). You want these cache dirs to reside on persistent volumes so that their contents are retained across multiple docker run invocations.

Interactive visualization

This release contains an interactive model visualization tool that can be used to explore various characteristics of a trained model. To start it, run:

python visualizer.py

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 StyleGAN3 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 StyleGAN3-T for AFHQv2 using 8 GPUs.
python train.py --outdir=~/training-runs --cfg=stylegan3-t --data=~/datasets/afhqv2-512x512.zip \
    --gpus=8 --batch=32 --gamma=8.2 --mirror=1

# Fine-tune StyleGAN3-R for MetFaces-U using 1 GPU, starting from the pre-trained FFHQ-U pickle.
python train.py --outdir=~/training-runs --cfg=stylegan3-r --data=~/datasets/metfacesu-1024x1024.zip \
    --gpus=8 --batch=32 --gamma=6.6 --mirror=1 --kimg=5000 --snap=5 \
    --resume=https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-r-ffhqu-1024x1024.pkl

# Train StyleGAN2 for FFHQ at 1024x1024 resolution using 8 GPUs.
python train.py --outdir=~/training-runs --cfg=stylegan2 --data=~/datasets/ffhq-1024x1024.zip \
    --gpus=8 --batch=32 --gamma=10 --mirror=1 --aug=noaug

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-stylegan3-t-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=eqt50k_int,eqr50k \
    --network=~/training-runs/00000-stylegan3-r-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=https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-t-ffhq-1024x1024.pkl

The first example looks up the training configuration and performs the same operation as if --metrics=eqt50k_int,eqr50k 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:

fid50k_full: Fréchet inception distance^[1] against the full dataset.
kid50k_full: Kernel inception distance^[2] against the full dataset.
pr50k3_full: Precision and recall^[3] againt the full dataset.
ppl2_wend: Perceptual path length^[4] in W, endpoints, full image.
eqt50k_int: Equivariance^[5] w.r.t. integer translation (EQ-T).
eqt50k_frac: Equivariance w.r.t. fractional translation (EQ-T_frac).
eqr50k: Equivariance w.r.t. rotation (EQ-R).

Legacy metrics:

fid50k: Fréchet inception distance against 50k real images.
kid50k: Kernel inception distance against 50k real images.
pr50k3: Precision and recall against 50k real images.
is50k: Inception score^[6] for CIFAR-10.

References:

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

Spectral analysis

The easiest way to inspect the spectral properties of a given generator is to use the built-in FFT mode in visualizer.py. In addition, you can visualize average 2D power spectra (Appendix A, Figure 15) as follows:

# Calculate dataset mean and std, needed in subsequent steps.
python avg_spectra.py stats --source=~/datasets/ffhq-1024x1024.zip

# Calculate average spectrum for the training data.
python avg_spectra.py calc --source=~/datasets/ffhq-1024x1024.zip \
    --dest=tmp/training-data.npz --mean=112.684 --std=69.509

# Calculate average spectrum for a pre-trained generator.
python avg_spectra.py calc \
    --source=https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-r-ffhq-1024x1024.pkl \
    --dest=tmp/stylegan3-r.npz --mean=112.684 --std=69.509 --num=70000

# Display results.
python avg_spectra.py heatmap tmp/training-data.npz
python avg_spectra.py heatmap tmp/stylegan3-r.npz
python avg_spectra.py slices tmp/training-data.npz tmp/stylegan3-r.npz

License

This work is made available under the Nvidia Source Code License.

Citation

@inproceedings{Karras2021,
  author = {Tero Karras and Miika Aittala and Samuli Laine and Erik H\"ark\"onen and Janne Hellsten and Jaakko Lehtinen and Timo Aila},
  title = {Alias-Free Generative Adversarial Networks},
  booktitle = {Proc. NeurIPS},
  year = {2021}
}

Development

This is a research reference implementation and is treated as a one-time code drop. As such, we do not accept outside code contributions in the form of pull requests.

Acknowledgements

We thank David Luebke, Ming-Yu Liu, Koki Nagano, Tuomas Kynkäänniemi, and Timo Viitanen for reviewing early drafts and helpful suggestions. Frédo Durand for early discussions. Tero Kuosmanen for maintaining our compute infrastructure. AFHQ authors for an updated version of their dataset. Getty Images for the training images in the Beaches dataset. We did not receive external funding or additional revenues for this project.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README.md

README.md

StyleGAN3-Fun
_{Let's have fun with StyleGAN2/ADA/3!}

Additions

Notebooks (Coming soon!)

Sponsors

Alias-Free Generative Adversarial Networks (StyleGAN3)
_{Official PyTorch implementation of the NeurIPS 2021 paper}

Release notes

Synthetic image detection

Additional material

Requirements

Getting started

Interactive visualization

Using networks from Python

Preparing datasets

Training

Quality metrics

Spectral analysis

License

Citation

Development

Acknowledgements

Files

README.md

Latest commit

History

README.md

File metadata and controls

StyleGAN3-FunLet's have fun with StyleGAN2/ADA/3!

Additions

Notebooks (Coming soon!)

Sponsors

Alias-Free Generative Adversarial Networks (StyleGAN3)Official PyTorch implementation of the NeurIPS 2021 paper

Release notes

Synthetic image detection

Additional material

Requirements

Getting started

Interactive visualization

Using networks from Python

Preparing datasets

Training

Quality metrics

Spectral analysis

License

Citation

Development

Acknowledgements

StyleGAN3-Fun
_{Let's have fun with StyleGAN2/ADA/3!}

Alias-Free Generative Adversarial Networks (StyleGAN3)
_{Official PyTorch implementation of the NeurIPS 2021 paper}