Guided image generation with conditional invertible neural networks

• Verfasst von: Ardizzone, Lynton [VerfasserIn]
• Verfasst von: Lüth, Carsten [VerfasserIn]
• Verfasst von: Kruse, Jakob [VerfasserIn]
• Verfasst von: Rother, Carsten [VerfasserIn]
• Verfasst von: Köthe, Ullrich [VerfasserIn]
• Titel: Guided image generation with conditional invertible neural networks
• Verf.angabe: Lynton Ardizzone, Carsten Lüth, Jakob Kruse, Carsten Rother, Ullrich Köthe
• E-Jahr: 2019
• Jahr: 10 Jul 2019
• Umfang: 11 S.
• Fussnoten: Gesehen am 19.07.2022
• Titel Quelle: Enthalten in: De.arxiv.org
• Ort Quelle: [S.l.] : Arxiv.org, 1991
• Jahr Quelle: 2019
• Band/Heft Quelle: (2019), Artikel-ID 1907.02392, Seite 1-11
• DOI: doi:10.48550/arXiv.1907.02392
• Datenträger: Online-Ressource
• Sprache: eng
• Sach-SW: 68T01
• Sach-SW: Computer Science - Computer Vision and Pattern Recognition
• Sach-SW: Computer Science - Machine Learning
• K10plus-PPN: 1810852498

Abstract

In this work, we address the task of natural image generation guided by a conditioning input. We introduce a new architecture called conditional invertible neural network (cINN). The cINN combines the purely generative INN model with an unconstrained feed-forward network, which efficiently preprocesses the conditioning input into useful features. All parameters of the cINN are jointly optimized with a stable, maximum likelihood-based training procedure. By construction, the cINN does not experience mode collapse and generates diverse samples, in contrast to e.g. cGANs. At the same time our model produces sharp images since no reconstruction loss is required, in contrast to e.g. VAEs. We demonstrate these properties for the tasks of MNIST digit generation and image colorization. Furthermore, we take advantage of our bi-directional cINN architecture to explore and manipulate emergent properties of the latent space, such as changing the image style in an intuitive way.