On Noise Injection in Generative Adversarial Networks
This work addresses a fundamental technique in GANs for improving image fidelity, but it appears incremental as it refines an existing method rather than introducing a new paradigm.
The paper tackles the unclear mechanism of noise injection in GANs by proposing a geometric framework based on Riemannian geometry to theoretically analyze it, finding that existing methods are incomplete and devising a new strategy that shows superiority in experiments on image generation and GAN inversion.
Noise injection has been proved to be one of the key technique advances in generating high-fidelity images. Despite its successful usage in GANs, the mechanism of its validity is still unclear. In this paper, we propose a geometric framework to theoretically analyze the role of noise injection in GANs. Based on Riemannian geometry, we successfully model the noise injection framework as fuzzy equivalence on the geodesic normal coordinates. Guided by our theories, we find that the existing method is incomplete and a new strategy for noise injection is devised. Experiments on image generation and GAN inversion demonstrate the superiority of our method.