Few-shot Image Generation with Elastic Weight Consolidation
This work provides an incremental improvement for researchers and practitioners in few-shot image generation, particularly for domains with extremely limited data.
This paper addresses few-shot image generation by adapting a pre-trained model from a large source domain to a target domain with very few examples. The method regularizes weight changes during adaptation to preserve source domain diversity while fitting target appearance, enabling high-quality generation even with fewer than 10 examples.
Few-shot image generation seeks to generate more data of a given domain, with only few available training examples. As it is unreasonable to expect to fully infer the distribution from just a few observations (e.g., emojis), we seek to leverage a large, related source domain as pretraining (e.g., human faces). Thus, we wish to preserve the diversity of the source domain, while adapting to the appearance of the target. We adapt a pretrained model, without introducing any additional parameters, to the few examples of the target domain. Crucially, we regularize the changes of the weights during this adaptation, in order to best preserve the information of the source dataset, while fitting the target. We demonstrate the effectiveness of our algorithm by generating high-quality results of different target domains, including those with extremely few examples (e.g., <10). We also analyze the performance of our method with respect to some important factors, such as the number of examples and the dissimilarity between the source and target domain.