Method p_sample

guided_diffusion/gaussian_diffusion.py:365–373

(self, model, x, t)

@register_sampler(name='ddpm')
class DDPM(SpacedDiffusion):
    def p_sample(self, model, x, t):
        out = self.p_mean_variance(model, x, t)
        sample = out['mean']

        noise = torch.randn_like(x)
        if t != 0:  # no noise when t == 0
            sample += torch.exp(0.5 * out['log_variance']) * noise

        return {'sample': sample, 'pred_xstart': out['pred_xstart']}
    

@register_sampler(name='ddim')