Research map / Image generation / GANs

Do WGANs succeed because they minimize the Wasserstein Distance? Lessons from Discrete Generators

ICLR 2025 · Image generation › GANs · approach: Stabilized GAN training

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Studies whether WGAN success stems from minimizing the Wasserstein distance, using discrete generators and theory to argue that successful WGANs do minimize a Wasserstein distance, but of a particular form.

Problem

Debate over whether WGANs' success in generating realistic images is due to minimizing the Wasserstein distance between generated and training distributions.

Approach

Argues, with theory and experiments, that successful WGANs do minimize the Wasserstein distance, but the form of the distance being minimized differs from the one usually assumed; analysis uses discrete generators.

Experimental setup

Builds on: WGAN

Closest papers by meaning

• A Two-Step Computation of the Exact GAN Wasserstein Distance (ICML 2018 · similarity 0.92)
• Banach Wasserstein GAN (NeurIPS 2018 · similarity 0.92)
• Improving the Improved Training of Wasserstein GANs: A Consistency Term and Its Dual Effect (ICLR 2018 · similarity 0.90)
• Towards Generalized Implementation of Wasserstein Distance in GANs (AAAI 2021 · similarity 0.89)
• Wasserstein of Wasserstein Loss for Learning Generative Models (ICML 2019 · similarity 0.89)
• Max-Sliced Wasserstein Distance and Its Use for GANs (CVPR 2019 · similarity 0.89)