Large Scale GAN Training for High Fidelity Natural Image Synthesis
This work addresses the challenge of high-fidelity natural image synthesis for applications in computer vision and generative modeling, representing a significant but incremental advance in scaling GANs.
The paper tackled the problem of generating high-resolution, diverse images from complex datasets like ImageNet by training Generative Adversarial Networks (GANs) at an unprecedented scale, achieving state-of-the-art results with an Inception Score of 166.5 and Frechet Inception Distance of 7.4 on ImageNet at 128x128 resolution.
Despite recent progress in generative image modeling, successfully generating high-resolution, diverse samples from complex datasets such as ImageNet remains an elusive goal. To this end, we train Generative Adversarial Networks at the largest scale yet attempted, and study the instabilities specific to such scale. We find that applying orthogonal regularization to the generator renders it amenable to a simple "truncation trick," allowing fine control over the trade-off between sample fidelity and variety by reducing the variance of the Generator's input. Our modifications lead to models which set the new state of the art in class-conditional image synthesis. When trained on ImageNet at 128x128 resolution, our models (BigGANs) achieve an Inception Score (IS) of 166.5 and Frechet Inception Distance (FID) of 7.4, improving over the previous best IS of 52.52 and FID of 18.6.