Function discretized_gaussian_log_likelihood

diff2flow/ddpm.py:533–559

Compute the log-likelihood of a Gaussian distribution discretizing to a given image. :param x: the target images. It is assumed that this was uint8 values, rescaled to the range [-1, 1]. :param means: the Gaussian mean Tensor. :param log_scales: the Gaussian log s

(x, *, means, log_scales)

def discretized_gaussian_log_likelihood(x, *, means, log_scales):
    """
    Compute the log-likelihood of a Gaussian distribution discretizing to a
    given image.

    :param x: the target images. It is assumed that this was uint8 values,
              rescaled to the range [-1, 1].
    :param means: the Gaussian mean Tensor.
    :param log_scales: the Gaussian log stddev Tensor.
    :return: a tensor like x of log probabilities (in nats).
    """
    assert x.shape == means.shape #== log_scales.shape      # TODO: is this valid?
    centered_x = x - means
    inv_stdv = torch.exp(-log_scales)
    plus_in = inv_stdv * (centered_x + 1.0 / 255.0)
    cdf_plus = approx_standard_normal_cdf(plus_in)
    min_in = inv_stdv * (centered_x - 1.0 / 255.0)
    cdf_min = approx_standard_normal_cdf(min_in)
    log_cdf_plus = torch.log(cdf_plus.clamp(min=1e-12))
    log_one_minus_cdf_min = torch.log((1.0 - cdf_min).clamp(min=1e-12))
    cdf_delta = cdf_plus - cdf_min
    log_probs = torch.where(
        x < -0.999, log_cdf_plus,
        torch.where(x > 0.999, log_one_minus_cdf_min, torch.log(cdf_delta.clamp(min=1e-12))),
    )
    assert log_probs.shape == x.shape
    return log_probs