Evaluating the distribution learning capabilities of GANs

Rege, Amit, Monteleoni, Claire

arXiv.org Machine Learning 

We find that by and large GANs fail to faithfully datasets. To our knowledge, the only instance of synthetic recreate point datasets which contain discontinous image datasets used for GAN evaluation have been to learn support or sharp bends with noise. Additionally, manifolds of convex polygons (specifically triangles) (Lucic on image datasets, we find that GANs do et al., 2018). Although, we also use polygons as a testbed not seem to learn to count the number of objects for our experiments, we focus on learning a manifold with of the same kind in an image. We also highlight multiple polygons where their number is fixed.

Duplicate Docs Excel Report

Title
None found

Similar Docs  Excel Report  more

TitleSimilaritySource
None found