Data-Efficient Instance Generation from Instance Discrimination
Ceyuan Yang† Yujun Shen‡ Yinghao Xu† Bolei Zhou† †The Chinese University of Hong Kong ‡ByteDance Inc.
Abstract
Generative Adversarial Networks (GANs) have significantly advanced image synthesis, however, the synthesis quality drops significantly given a limited amount of training data. To improve the data efficiency of GAN training, prior work typically employs data augmentation to mitigate the overfitting of the discriminator yet still learn the discriminator with a bi-classification (i.e., real vs. fake) task. In this work, we propose a data-efficient Instance Generation (InsGen) method based on instance discrimination. Concretely, besides differentiating the real domain from the fake domain, the discriminator is required to distinguish every individual image, no matter it comes from the training set or from the generator. In this way, the discriminator can benefit from the infinite synthesized samples for training, alleviating the overfitting problem caused by insufficient training data. A noise perturbation strategy is further introduced to improve its discriminative power. Meanwhile, the learned instance discrimination capability from the discriminator is in turn exploited to encourage the generator for diverse generation. Extensive experiments demonstrate the effectiveness of our meth
中文速览
小数据场景下训练生成对抗网络（GAN）时，判别器因样本太少容易过拟合、失去辨别力，导致生成图像质量大幅下降。为此，本文提出了一种名为 InsGen 的方法：除了让判别器完成传统的"真图 vs. 假图"二分类任务外，还额外要求它把每一张图像——无论真实还是生成的——都当作独立个体来区分（实例判别，instance discrimination）；由于生成器可以源源不断地采样，这相当于给判别器提供了近乎无限的训练样本，从根本上缓解了过拟合。同时，作者引入噪声扰动策略来进一步提升判别难度，并将判别器学到的实例区分能力反馈给生成器，促使它生成更多样化的图像。实验表明，在仅用 2000 张 FFHQ 人脸图像的极低数据量设置下，InsGen 将 FID 指标从 15.60 降至 11.92，相比当时最优方法提升了 23.5%，证明该方法在多种数据规模下都能显著改善生成质量与多样性。
原文 arXiv:2106.04566;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2106.04566v1