Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper addresses intra-mode collapse in GANs, where multiple latent variables map to similar outputs, by introducing a pairing regularizer that enforces local consistency between latent variables and generated samples. The effect of the regularizer varies based on the training regime: it encourages exploration in collapse-prone scenarios and refines data density in stabilized training. Experiments on toy distributions and real images show the regularizer complements existing stabilization techniques, improving coverage and precision.
GANs can be made more robust by explicitly discouraging redundant mappings from latent space, leading to higher precision without sacrificing recall.
Mode collapse remains a fundamental challenge in training generative adversarial networks (GANs). While existing works have primarily focused on inter-mode collapse, such as mode dropping, intra-mode collapse-where many latent variables map to the same or highly similar outputs-has received significantly less attention. In this work, we propose a pairing regularizer jointly optimized with the generator to mitigate the many-to-one collapse by enforcing local consistency between latent variables and generated samples. We show that the effect of pairing regularization depends on the dominant failure mode of training. In collapse-prone regimes with limited exploration, pairing encourages structured local exploration, leading to improved coverage and higher recall. In contrast, under stabilized training with sufficient exploration, pairing refines the generator's induced data density by discouraging redundant mappings, thereby improving precision without sacrificing recall. Extensive experiments on both toy distributions and real-image benchmarks demonstrate that the proposed regularizer effectively complements existing stabilization techniques by directly addressing intra-mode collapse.