Method q_sample

diff2flow/ddpm.py:199–209

Diffuse the data for a given number of diffusion steps => sample x_t ~ q(x_t | x_0) with x_t = sqrt(alpha_bar_t) * x_0 + sqrt(1 - alpha_bar_t) * noise

(self, x_start, t, noise=None)

        return mean, variance, log_variance
    
    def q_sample(self, x_start, t, noise=None):
        """
        Diffuse the data for a given number of diffusion steps
                => sample x_t ~ q(x_t | x_0)    with
        x_t = sqrt(alpha_bar_t) * x_0 + sqrt(1 - alpha_bar_t) * noise
        """
        if noise is None:
            noise = torch.randn_like(x_start)
        sqrt_alpha_cumprod_t = extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape)
        sqrt_one_minus_alpha_cumprod_t = extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape)
        return sqrt_alpha_cumprod_t * x_start + sqrt_one_minus_alpha_cumprod_t * noise
    
    def get_v(self, x_start, noise, t):
        """ v-parameterization: sqrt(alpha_bar_t) * eps - sqrt(1 - alpha_bar_t) * x_0 """