Generative adversarial interpolative autoencoding: adversarial training on latent space interpolations encourage convex latent distributions

By Tim Sainburg, Marvin Thielk, Bradley H. Theilman, Benjamin Migliori, Timothy Q. Gentner. 2018

56 citations recorded by the source corpus.

We present a neural network architecture based upon the Autoencoder (AE) and Generative Adversarial Network (GAN) that promotes a convex latent distribution by training adversarially on latent space interpolations. By using an AE as both the generator and discriminator of a GAN, we pass a pixel-wise error function across the discriminator, yielding an AE which produces non-blurry samples that match both high- and low-level features of the original images. Interpolations between images in this space remain within the latent-space distribution of real images as trained by the discriminator, and therfore preserve realistic resemblances to the network inputs. Code available at this https URL

Subfields: Computer Vision and Pattern Recognition, Media Technology, Generative Models, Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Xiaogang Wang
2. Luc Van Gool
3. Dacheng Tao
4. Ming-Hsuan Yang
5. Trevor Darrell
6. Andrew Zisserman
7. Yi Yang
8. Bernt Schiele

In Media Technology

1. Xinyu Gong
2. Yoshua Bengio
3. Timo Aila
4. C. Pal
5. Yu Liu
6. Tero Karras
7. Xiaogang Wang
8. Samuli Laine

In Generative Models

1. Stefano Ermon
2. Jun-Yan Zhu
3. Richard Zhang
4. Tero Karras
5. Timo Aila
6. Samuli Laine
7. Eli Shechtman
8. Yang Song

In Representation & Self-Supervised Learning

1. Yoshua Bengio
2. Xiaogang Wang
3. R. Devon Hjelm
4. Vikas Verma
5. C. Pal
6. Alex Lamb
7. Timo Aila
8. Quoc V. Le

Citations in the corpus

• A Style-Based Generator Architecture for Generative Adversarial Networks (2018, background): “The work on generator side has mostly focused on the exact distribution in the input latent space [4] or shaping the input latent space via Gaussian mixture models [3], clustering [44], or encouraging convexity [48].”
• Conditional Adversarial Generative Flow for Controllable Image Synthesis (2019, background): “Many following works improve generative adversarial networks by better loss, training skills and evaluating metrics [1, 13, 19, 34, 16, 24, 27, 35].”
• On Adversarial Mixup Resynthesis (2019, methodology): “spectral normalization (Miyato et al., 2018) to impose a Lipschitz constraint on the discriminator, which is known to be very effective in minimising stability issues in training. The GAIA algorithm (Sainburg et al., 2018) uses a BEGAN framework with an additional interpolation-based adversarial obj”