Adaptive Feature Interpolation for Low-Shot Image Generation
This addresses the challenge of few-shot image generation for researchers and practitioners, offering a novel unsupervised approach that is incremental in improving training stability.
The paper tackles the problem of training generative models, particularly GANs, in low-data settings by proposing an implicit data augmentation method that stabilizes training and synthesizes high-quality samples without labels, showing significant improvements over baselines with hundreds of training samples.
Training of generative models especially Generative Adversarial Networks can easily diverge in low-data setting. To mitigate this issue, we propose a novel implicit data augmentation approach which facilitates stable training and synthesize high-quality samples without need of label information. Specifically, we view the discriminator as a metric embedding of the real data manifold, which offers proper distances between real data points. We then utilize information in the feature space to develop a fully unsupervised and data-driven augmentation method. Experiments on few-shot generation tasks show the proposed method significantly improve results from strong baselines with hundreds of training samples.