Domain Partitioning Network

TL;DR

The paper introduces DoPaNet, a novel adversarial training framework with multiple discriminators and a classifier to effectively address mode collapse and improve distribution coverage in generative models.

Contribution

It proposes a new method with multiple discriminators and a classifier to prevent mode collapse and enhance distribution coverage in GANs.

Findings

DoPaNet covers the real data distribution more effectively.

It outperforms existing methods in experiments.

It allows control over generated modes.

Abstract

Standard adversarial training involves two agents, namely a generator and a discriminator, playing a mini-max game. However, even if the players converge to an equilibrium, the generator may only recover a part of the target data distribution, in a situation commonly referred to as mode collapse. In this work, we present the Domain Partitioning Network (DoPaNet), a new approach to deal with mode collapse in generative adversarial learning. We employ multiple discriminators, each encouraging the generator to cover a different part of the target distribution. To ensure these parts do not overlap and collapse into the same mode, we add a classifier as a third agent in the game. The classifier decides which discriminator the generator is trained against for each sample. Through experiments on toy examples and real images, we show the merits of DoPaNet in covering the real distribution and its superiority with respect to the competing methods. Besides, we also show that we can control the modes from which samples are generated using DoPaNet.