GANSpace: Discovering Interpretable GAN Controls

Figure 1: Sequences of image edits performed using control discovered with our method, applied to three different GANs. The white insets specify the particular edits using notation explained in Section 3.4 ('Layer-wise Edits').

GANSpace: Discovering Interpretable GAN Controls
Erik Härkönen^1,2, Aaron Hertzmann², Jaakko Lehtinen^1,3, Sylvain Paris^2
1Aalto University, ²Adobe Research, ³NVIDIA
https://arxiv.org/abs/2004.02546

Abstract: This paper describes a simple technique to analyze Generative Adversarial Networks (GANs) and create interpretable controls for image synthesis, such as change of viewpoint, aging, lighting, and time of day. We identify important latent directions based on Principal Components Analysis (PCA) applied in activation space. Then, we show that interpretable edits can be defined based on layer-wise application of these edit directions. Moreover, we show that BigGAN can be controlled with layer-wise inputs in a StyleGAN-like manner. A user may identify a large number of interpretable controls with these mechanisms. We demonstrate results on GANs from various datasets.

Video: https://youtu.be/jdTICDa_eAI

Setup

See the setup instructions.

Usage

This repository includes versions of BigGAN, StyleGAN, and StyleGAN2 modified to support per-layer latent vectors.

Interactive model exploration

# Explore BigGAN-deep husky
python interactive.py --model=BigGAN-512 --class=husky --layer=generator.gen_z -n=1_000_000

# Explore StyleGAN2 ffhq in W space
python interactive.py --model=StyleGAN2 --class=ffhq --layer=style --use_w -n=1_000_000 -b=10_000

# Explore StyleGAN2 cars in Z space
python interactive.py --model=StyleGAN2 --class=car --layer=style -n=1_000_000 -b=10_000

# Apply previously saved edits interactively
python interactive.py --model=StyleGAN2 --class=ffhq --layer=style --use_w --inputs=out/directions

Visualize principal components

# Visualize StyleGAN2 ffhq W principal components
python visualize.py --model=StyleGAN2 --class=ffhq --use_w --layer=style -b=10_000

# Create videos of StyleGAN wikiart components (saved to ./out)
python visualize.py --model=StyleGAN --class=wikiart --use_w --layer=g_mapping -b=10_000 --batch --video

Options

Command line paramaters:
  --model      one of [ProGAN, BigGAN-512, BigGAN-256, BigGAN-128, StyleGAN, StyleGAN2]
  --class      class name; leave empty to list options
  --layer      layer at which to perform PCA; leave empty to list options
  --use_w      treat W as the main latent space (StyleGAN / StyleGAN2)
  --inputs     load previously exported edits from directory
  --sigma      number of stdevs to use in visualize.py
  -n           number of PCA samples
  -b           override automatic minibatch size detection
  -c           number of components to keep

Reproducibility

All figures presented in the main paper can be recreated using the included Jupyter notebooks:

• Figure 1: figure_teaser.ipynb
• Figure 2: figure_pca_illustration.ipynb
• Figure 3: figure_pca_cleanup.ipynb
• Figure 4: figure_style_content_sep.ipynb
• Figure 5: figure_supervised_comp.ipynb
• Figure 6: figure_biggan_style_resampling.ipynb
• Figure 7: figure_edit_zoo.ipynb

Known issues

• The interactive viewer sometimes freezes on startup on Ubuntu 18.04. The freeze is resolved by clicking on the terminal window and pressing the control key. Any insight into the issue would be greatly appreciated!

Integrating a new model

1. Create a wrapper for the model in models/wrappers.py using the BaseModel interface.
2. Add the model to get_model() in models/wrappers.py.

Importing StyleGAN checkpoints from TensorFlow

It is possible to import trained StyleGAN and StyleGAN2 weights from TensorFlow into GANSpace.

StyleGAN

1. Install TensorFlow: conda install tensorflow-gpu=1.*.
2. Modify methods __init__(), load_model() in models/wrappers.py under class StyleGAN.

StyleGAN2

1. Follow the instructions here. Make sure to use the forked repository in the conversion for compatibility reasons.
2. Save the converted checkpoint as checkpoints/stylegan2/<dataset>_<resolution>.pt.
3. Modify methods __init__(), download_checkpoint() in models/wrappers.py under class StyleGAN2.

Acknowledgements

We would like to thank:

• The authors of the PyTorch implementations of BigGAN, StyleGAN, and StyleGAN2:
  Thomas Wolf, Piotr Bialecki, Thomas Viehmann, and Kim Seonghyeon.
• Joel Simon from ArtBreeder for providing us with the landscape model for StyleGAN.
  (unfortunately we cannot distribute this model)
• David Bau and colleagues for the excellent GAN Dissection project.
• Justin Pinkney for the Awesome Pretrained StyleGAN collection.
• Tuomas Kynkäänniemi for giving us a helping hand with the experiments.
• The Aalto Science-IT project for providing computational resources for this project.

License

The code of this repository is released under the Apache 2.0 license.
The directory netdissect is a derivative of the GAN Dissection project, and is provided under the MIT license.
The directories models/biggan and models/stylegan2 are provided under the MIT license.