Is the noise_strength variable used as a augmentation?

[hitsz-zuoqi]

    def sample_loss(self, x0, noise=None):
        if noise is None:
            noise = torch.randn_like(x0)
            if self.noise_strength > 0:
                b, c, f, _, _= x0.shape
                offset_noise = torch.randn(b, c, f, 1, 1, device=x0.device)
                noise = noise + self.noise_strength * offset_noise
        return noise

[Steven-SWZhang]

Hello, here we employ the offset noise to enhance the quality of the videos, as a common training method we use.