GAN-Control: Explicitly Controllable GANs

Alon Shoshan
Nadav Bhonker
Igor Kviatkovsky
Gerard Medioni

Amazon One

arXiv [Paper]
[Supplementary]

# Abstract

Proposed framework: In Phase 1, we construct every batch so that for each attribute, there is a pair of latent vectors sharing a corresponding sub-vector, \(\mathbf{z}^k\). In addition to the adversarial loss, each image in the batch is compared in a contrastive manner, attribute-by-attribute, to all others, taking into account if it has the same or a different sub-vector. In Phase 2, encoders are trained to map interpretable parameters to suitable latent vectors. Inference: An explicit control over the attribute \(k\) is achieved by setting the \(k\)th encoder input to a required value.

# Applications

Illum. 1 | Illum. 2 | Illum. 3 | Illum. 4 | Illum. 5 |

Yaw=\(30^\text{o}\) | \(15^\text{o}\) | \(0^\text{o}\) | \(-15^\text{o}\) | \(-30^\text{o}\) |

Exp. 1 | Exp. 2 | Exp. 3 | Exp. 4 | Exp. 5 |

Color 1 | Color 2 | Color 3 | Color 4 | Color 5 |

Age=\(15\)yo | \(30\)yo | \(45\)yo | \(60\)yo | \(75\)yo |

Yaw=\(30^\text{o}\) | \(15^\text{o}\) | \(0^\text{o}\) | \(-15^\text{o}\) | \(-30^\text{o}\) |

Age=\(15\)yo | \(30\)yo | \(45\)yo | \(60\)yo | \(75\)yo |

Exp. 1 | Exp. 2 | Exp. 3 | Exp. 4 | Exp. 5 |

Style. 1 | Style. 2 | Style. 3 | Style. 4 | Style. 5 |

Yaw=\(30^\text{o}\) | \(15^\text{o}\) | \(0^\text{o}\) | \(-15^\text{o}\) | \(-30^\text{o}\) |

Roll=\(20^\text{o}\) | \(-20^\text{o}\) | \(0^\text{o}\) | Pitch=\(-10^\text{o}\) | \(10^\text{o}\) |

Input [1] |
Projected |
Yaw=\(30^\text{o}\) Pitch=\(0^\text{o}\) |
Yaw=\(25^\text{o}\) Pitch=\(-15^\text{o}\) |
Yaw=\(-25^\text{o}\) Pitch=\(15^\text{o}\) |

Input [2] | Projected | Right | Front | Left |

Input [3] | Projected | Age=\(15\)yo | \(45\)yo | \(70\)yo |

Input [4] | Projected | Exp. 1 | Exp. 2 | Exp. 3 |

[2] The original image is at www.flickr.com/photos/dereknolan/5309847731 and is licensed under: http://www.creativecommons.org/licenses/by/2.0

[3] The original image is at www.flickr.com/photos/67548743@N02/6854926480 and is licensed under: http://www.creativecommons.org/licenses/by/2.0

[4] The original image is at www.flickr.com/photos/ugacommunications/6005899336 and is licensed under: http://www.creativecommons.org/licenses/by-nc/2.0

# GAN-Control: Explicitly Controllable GANs

# Citation

@misc{shoshan2021gancontrol, title={GAN-Control: Explicitly Controllable GANs}, author={Alon Shoshan and Nadav Bhonker and Igor Kviatkovsky and Gerard Medioni}, year={2021}, eprint={2101.02477}, archivePrefix={arXiv}, primaryClass={cs.CV} }