MearcAll papers

arXiv 1812.04948 · cs.NE cs.LG stat.ML

A Style-Based Generator Architecture for Generative Adversarial Networks

Tero Karras, Samuli Laine, Timo Aila

Claims were extracted automatically by reducer-1 on 2026-11-12. The paper's license does not allow republishing it, so long claims are shortened to a quotation. Read the full text on arXiv.

No reproduction recordedclm_5808d389
We propose an alternative generator architecture for generative adversarial networks, borrowing from style transfer literature.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_677a50ea
The new generator improves the state-of-the-art in terms of traditional distribution quality metrics, leads to demonstrably better interpolation properties, and also better disentangles the latent factors of variation.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_c66d2783
To quantify interpolation quality and disentanglement, we propose two new, automated methods that are applicable to any generator architecture.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_4a0f3fbd
Finally, we introduce a new, highly varied and high-quality dataset of human faces.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_cd557892
Motivated by style transfer literature [27], we re-design the generator architecture in a way that exposes novel ways to control the image synthesis process.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_11bb2606
The input latent space must follow the probability density of the training data, and we argue that this leads to some degree of unavoidable entanglement.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_3ed7cc32
Using these metrics, we show that compared to a traditional generator architecture, our generator admits a more linear, less entangled representation of different factors of variation.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_0358ee13
Finally, we present a new dataset of human faces (Flickr-Faces-HQ, FFHQ) that offers much higher quality and covers considerably wider variation than existing high-resolution datasets (Appendix A).

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_34aa8442
Before studying the properties of our generator, we demonstrate experimentally that the redesign does not compromise image quality but, in fact, improves it considerably.

Quality of generated images · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_50bc47ec
We first switch to an improved baseline (B) by using bilinear up/downsampling operations [64], longer training, and tuned hyperparameters.

Quality of generated images · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_f108b174
Finally, we introduce the noise inputs ( E) that improve the results further, as well as novelmixing regularization(F) that decorrelates neighboring styles and enables more finegrained control over the generated … (excerpt)

Quality of generated images · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_002ca2a5
We observe that the style-based generator ( E) improves FIDs quite significantly over the traditional generator ( B), almost 20%, corroborating the large-scale ImageNet measurements made in parallel work [6, … (excerpt)

Quality of generated images · page 3 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_5d95663b
We can view the mapping network and affine transformations as a way to draw samples for each style from a learned distribution, and the synthesis network as a way to … (excerpt)

Properties of the style-based generator · page 3 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_f3a31593
We propose another metric that quantifies this effect by measuring how well the latent-space points can be separated into two distinct sets via a linear hyperplane, so that each set … (excerpt)

Linear separability · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_ba6bb5da
This shows that even the traditional generator architecture performs better when we introduce an intermediate latent space that does not have to follow the distribution of the training data.

Linear separability · page 8 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_4609f3a5
This is true in terms of established quality metrics, and we further believe that our investigations to the separation of high-level attributes and stochastic effects, as well as the linearity … (excerpt)

Conclusion · page 8 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_450e9568
For our improved baseline (B in Table 1), we make several modifications to improve the overall result quality.

Hyperparameters and training details · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_7a79c76d
Even when the training has reached the full 10242 resolution, the slowly rising path lengths indicate that the improvements in FID come at the cost of a more entangled representation.

Training convergence · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

Reproduced one of these claims?

Record your result in the Mearc app on your own machine, then use Publish this result with this paper's id (1812.04948) and the claim id shown next to the claim. Mearc creates a file that holds only what you choose to publish. Add it to the public claim data by pull request; a maintainer other than you reviews it before it counts.