clm_5808d389We propose an alternative generator architecture for generative adversarial networks, borrowing from style transfer literature.
arXiv 1812.04948 · cs.NE cs.LG stat.ML
Read on arXiv · arXiv non-exclusive license
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.
clm_5808d389We propose an alternative generator architecture for generative adversarial networks, borrowing from style transfer literature.
clm_677a50eaThe 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.
clm_c66d2783To quantify interpolation quality and disentanglement, we propose two new, automated methods that are applicable to any generator architecture.
clm_4a0f3fbdFinally, we introduce a new, highly varied and high-quality dataset of human faces.
clm_cd557892Motivated by style transfer literature [27], we re-design the generator architecture in a way that exposes novel ways to control the image synthesis process.
clm_11bb2606The input latent space must follow the probability density of the training data, and we argue that this leads to some degree of unavoidable entanglement.
clm_3ed7cc32Using 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.
clm_0358ee13Finally, 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).
clm_34aa8442Before studying the properties of our generator, we demonstrate experimentally that the redesign does not compromise image quality but, in fact, improves it considerably.
res_50bc47ecWe first switch to an improved baseline (B) by using bilinear up/downsampling operations [64], longer training, and tuned hyperparameters.
res_f108b174Finally, 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)
res_002ca2a5We 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)
clm_5d95663bWe 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)
clm_f3a31593We 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)
clm_ba6bb5daThis 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.
clm_4609f3a5This 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)
res_450e9568For our improved baseline (B in Table 1), we make several modifications to improve the overall result quality.
res_7a79c76dEven 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.
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.