KarrasVeScheduler
Last updated
Last updated
KarrasVeScheduler
is a stochastic sampler tailored o variance-expanding (VE) models. It is based on the and papers.
( sigma_min: float = 0.02sigma_max: float = 100s_noise: float = 1.007s_churn: float = 80s_min: float = 0.05s_max: float = 50 )
Parameters
sigma_min (float
, defaults to 0.02) β The minimum noise magnitude.
sigma_max (float
, defaults to 100) β The maximum noise magnitude.
s_noise (float
, defaults to 1.007) β The amount of additional noise to counteract loss of detail during sampling. A reasonable range is [1.000, 1.011].
s_churn (float
, defaults to 80) β The parameter controlling the overall amount of stochasticity. A reasonable range is [0, 100].
s_min (float
, defaults to 0.05) β The start value of the sigma range to add noise (enable stochasticity). A reasonable range is [0, 10].
s_max (float
, defaults to 50) β The end value of the sigma range to add noise. A reasonable range is [0.2, 80].
A stochastic scheduler tailored to variance-expanding models.
This model inherits from and . Check the superclass documentation for the generic methods the library implements for all schedulers such as loading and saving.
add_noise_to_input
( sample: FloatTensorsigma: floatgenerator: typing.Optional[torch._C.Generator] = None )
Parameters
sample (torch.FloatTensor
) β The input sample.
sigma (float
) β
generator (torch.Generator
, optional) β A random number generator.
Explicit Langevin-like βchurnβ step of adding noise to the sample according to a gamma_i β₯ 0
to reach a higher noise level sigma_hat = sigma_i + gamma_i*sigma_i
.
scale_model_input
( sample: FloatTensortimestep: typing.Optional[int] = None ) β torch.FloatTensor
Parameters
sample (torch.FloatTensor
) β The input sample.
timestep (int
, optional) β The current timestep in the diffusion chain.
Returns
torch.FloatTensor
A scaled input sample.
Ensures interchangeability with schedulers that need to scale the denoising model input depending on the current timestep.
set_timesteps
( num_inference_steps: intdevice: typing.Union[str, torch.device] = None )
Parameters
num_inference_steps (int
) β The number of diffusion steps used when generating samples with a pre-trained model.
device (str
or torch.device
, optional) β The device to which the timesteps should be moved to. If None
, the timesteps are not moved.
Sets the discrete timesteps used for the diffusion chain (to be run before inference).
step
( model_output: FloatTensorsigma_hat: floatsigma_prev: floatsample_hat: FloatTensorreturn_dict: bool = True ) β ~schedulers.scheduling_karras_ve.KarrasVESchedulerOutput
or tuple
Parameters
model_output (torch.FloatTensor
) β The direct output from learned diffusion model.
sigma_hat (float
) β
sigma_prev (float
) β
sample_hat (torch.FloatTensor
) β
return_dict (bool
, optional, defaults to True
) β Whether or not to return a ~schedulers.scheduling_karras_ve.KarrasVESchedulerOutput
or tuple
.
Returns
~schedulers.scheduling_karras_ve.KarrasVESchedulerOutput
or tuple
If return_dict is True
, ~schedulers.scheduling_karras_ve.KarrasVESchedulerOutput
is returned, otherwise a tuple is returned where the first element is the sample tensor.
Predict the sample from the previous timestep by reversing the SDE. This function propagates the diffusion process from the learned model outputs (most often the predicted noise).
step_correct
( model_output: FloatTensorsigma_hat: floatsigma_prev: floatsample_hat: FloatTensorsample_prev: FloatTensorderivative: FloatTensorreturn_dict: bool = True ) β prev_sample (TODO)
Parameters
model_output (torch.FloatTensor
) β The direct output from learned diffusion model.
sigma_hat (float
) β TODO
sigma_prev (float
) β TODO
sample_hat (torch.FloatTensor
) β TODO
sample_prev (torch.FloatTensor
) β TODO
derivative (torch.FloatTensor
) β TODO
Returns
prev_sample (TODO)
updated sample in the diffusion chain. derivative (TODO): TODO
Corrects the predicted sample based on the model_output
of the network.
( prev_sample: FloatTensorderivative: FloatTensorpred_original_sample: typing.Optional[torch.FloatTensor] = None )
Parameters
prev_sample (torch.FloatTensor
of shape (batch_size, num_channels, height, width)
for images) β Computed sample (x_{t-1}) of previous timestep. prev_sample
should be used as next model input in the denoising loop.
derivative (torch.FloatTensor
of shape (batch_size, num_channels, height, width)
for images) β Derivative of predicted original image sample (x_0).
pred_original_sample (torch.FloatTensor
of shape (batch_size, num_channels, height, width)
for images) β The predicted denoised sample (x_{0}) based on the model output from the current timestep. pred_original_sample
can be used to preview progress or for guidance.
Output class for the schedulerβs step function output.
For more details on the parameters, see . The grid search values used to find the optimal {s_noise, s_churn, s_min, s_max}
for a specific model are described in Table 5 of the paper.
return_dict (bool
, optional, defaults to True
) β Whether or not to return a or tuple
.