280 lines
14 KiB
Python
280 lines
14 KiB
Python
"""SAMPLING ONLY."""
|
|
|
|
import torch
|
|
import numpy as np
|
|
from tqdm import tqdm
|
|
|
|
from modules.shared import state
|
|
from modules.sd_samplers_common import InterruptedException
|
|
|
|
from videocrafter.lvdm.models.modules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like
|
|
|
|
|
|
class DDIMSampler(object):
|
|
def __init__(self, model, schedule="linear", **kwargs):
|
|
super().__init__()
|
|
self.model = model
|
|
self.ddpm_num_timesteps = model.num_timesteps
|
|
self.schedule = schedule
|
|
self.counter = 0
|
|
self.noise_gen = torch.Generator(device='cpu')
|
|
|
|
def register_buffer(self, name, attr):
|
|
if type(attr) == torch.Tensor:
|
|
if attr.device != torch.device("cuda"):
|
|
attr = attr.to(torch.device("cuda"))
|
|
setattr(self, name, attr)
|
|
|
|
def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True):
|
|
self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps,
|
|
num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose)
|
|
alphas_cumprod = self.model.alphas_cumprod
|
|
assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep'
|
|
to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device)
|
|
|
|
self.register_buffer('betas', to_torch(self.model.betas))
|
|
self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
|
|
self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev))
|
|
|
|
# calculations for diffusion q(x_t | x_{t-1}) and others
|
|
self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu())))
|
|
self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu())))
|
|
self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu())))
|
|
self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu())))
|
|
self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1)))
|
|
|
|
# ddim sampling parameters
|
|
ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(),
|
|
ddim_timesteps=self.ddim_timesteps,
|
|
eta=ddim_eta,verbose=verbose)
|
|
self.register_buffer('ddim_sigmas', ddim_sigmas)
|
|
self.register_buffer('ddim_alphas', ddim_alphas)
|
|
self.register_buffer('ddim_alphas_prev', ddim_alphas_prev)
|
|
self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas))
|
|
sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
|
|
(1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * (
|
|
1 - self.alphas_cumprod / self.alphas_cumprod_prev))
|
|
self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps)
|
|
|
|
@torch.no_grad()
|
|
def sample(self,
|
|
S,
|
|
batch_size,
|
|
shape,
|
|
conditioning=None,
|
|
callback=None,
|
|
img_callback=None,
|
|
quantize_x0=False,
|
|
eta=0.,
|
|
mask=None,
|
|
x0=None,
|
|
temperature=1.,
|
|
noise_dropout=0.,
|
|
score_corrector=None,
|
|
corrector_kwargs=None,
|
|
verbose=True,
|
|
schedule_verbose=False,
|
|
x_T=None,
|
|
log_every_t=100,
|
|
unconditional_guidance_scale=1.,
|
|
unconditional_conditioning=None,
|
|
postprocess_fn=None,
|
|
sample_noise=None,
|
|
cond_fn=None,
|
|
# this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
|
|
**kwargs
|
|
):
|
|
|
|
# check condition bs
|
|
if conditioning is not None:
|
|
if isinstance(conditioning, dict):
|
|
try:
|
|
cbs = conditioning[list(conditioning.keys())[0]].shape[0]
|
|
if cbs != batch_size:
|
|
print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
|
|
except:
|
|
# cbs = conditioning[list(conditioning.keys())[0]][0].shape[0]
|
|
pass
|
|
else:
|
|
if conditioning.shape[0] != batch_size:
|
|
print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
|
|
|
|
self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=schedule_verbose)
|
|
|
|
# make shape
|
|
if len(shape) == 3:
|
|
C, H, W = shape
|
|
size = (batch_size, C, H, W)
|
|
elif len(shape) == 4:
|
|
C, T, H, W = shape
|
|
size = (batch_size, C, T, H, W)
|
|
|
|
samples, intermediates = self.ddim_sampling(conditioning, size,
|
|
callback=callback,
|
|
img_callback=img_callback,
|
|
quantize_denoised=quantize_x0,
|
|
mask=mask, x0=x0,
|
|
ddim_use_original_steps=False,
|
|
noise_dropout=noise_dropout,
|
|
temperature=temperature,
|
|
score_corrector=score_corrector,
|
|
corrector_kwargs=corrector_kwargs,
|
|
x_T=x_T,
|
|
log_every_t=log_every_t,
|
|
unconditional_guidance_scale=unconditional_guidance_scale,
|
|
unconditional_conditioning=unconditional_conditioning,
|
|
postprocess_fn=postprocess_fn,
|
|
sample_noise=sample_noise,
|
|
cond_fn=cond_fn,
|
|
verbose=verbose,
|
|
**kwargs
|
|
)
|
|
return samples, intermediates
|
|
|
|
@torch.no_grad()
|
|
def ddim_sampling(self, cond, shape,
|
|
x_T=None, ddim_use_original_steps=False,
|
|
callback=None, timesteps=None, quantize_denoised=False,
|
|
mask=None, x0=None, img_callback=None, log_every_t=100,
|
|
temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
|
|
unconditional_guidance_scale=1., unconditional_conditioning=None,
|
|
postprocess_fn=None,sample_noise=None,cond_fn=None,
|
|
uc_type=None, verbose=True, **kwargs,
|
|
):
|
|
|
|
device = self.model.betas.device
|
|
|
|
b = shape[0]
|
|
if x_T is None:
|
|
img = torch.randn(shape, device=device)
|
|
else:
|
|
img = x_T
|
|
|
|
if timesteps is None:
|
|
timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps
|
|
elif timesteps is not None and not ddim_use_original_steps:
|
|
subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1
|
|
timesteps = self.ddim_timesteps[:subset_end]
|
|
intermediates = {'x_inter': [img], 'pred_x0': [img]}
|
|
time_range = reversed(range(0,timesteps)) if ddim_use_original_steps else np.flip(timesteps)
|
|
total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
|
|
if verbose:
|
|
iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps)
|
|
else:
|
|
iterator = time_range
|
|
|
|
state.sampling_steps = total_steps
|
|
|
|
for i, step in enumerate(iterator):
|
|
state.sampling_step = i
|
|
if state.interrupted:
|
|
raise InterruptedException
|
|
|
|
index = total_steps - i - 1
|
|
ts = torch.full((b,), step, device=device, dtype=torch.long)
|
|
|
|
if postprocess_fn is not None:
|
|
img = postprocess_fn(img, ts)
|
|
|
|
outs = self.p_sample_ddim(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps,
|
|
quantize_denoised=quantize_denoised, temperature=temperature,
|
|
noise_dropout=noise_dropout, score_corrector=score_corrector,
|
|
corrector_kwargs=corrector_kwargs,
|
|
unconditional_guidance_scale=unconditional_guidance_scale,
|
|
unconditional_conditioning=unconditional_conditioning,
|
|
sample_noise=sample_noise,cond_fn=cond_fn,uc_type=uc_type, **kwargs,)
|
|
img, pred_x0 = outs
|
|
|
|
if mask is not None:
|
|
# use mask to blend x_known_t-1 & x_sample_t-1
|
|
assert x0 is not None
|
|
x0 = x0.to(img.device)
|
|
mask = mask.to(img.device)
|
|
t = torch.tensor([step-1]*x0.shape[0], dtype=torch.long, device=img.device)
|
|
img_known = self.model.q_sample(x0, t)
|
|
img = img_known * mask + (1. - mask) * img
|
|
|
|
if callback: callback(i)
|
|
if img_callback: img_callback(pred_x0, i)
|
|
|
|
if index % log_every_t == 0 or index == total_steps - 1:
|
|
intermediates['x_inter'].append(img)
|
|
intermediates['pred_x0'].append(pred_x0)
|
|
if state.skipped:
|
|
break
|
|
|
|
return img, intermediates
|
|
|
|
@torch.no_grad()
|
|
def p_sample_ddim(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
|
|
temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
|
|
unconditional_guidance_scale=1., unconditional_conditioning=None, sample_noise=None,
|
|
cond_fn=None, uc_type=None,
|
|
**kwargs,
|
|
):
|
|
b, *_, device = *x.shape, x.device
|
|
if x.dim() == 5:
|
|
is_video = True
|
|
else:
|
|
is_video = False
|
|
if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
|
|
e_t = self.model.apply_model(x, t, c, **kwargs) # unet denoiser
|
|
else:
|
|
# with unconditional condition
|
|
if isinstance(c, torch.Tensor):
|
|
e_t = self.model.apply_model(x, t, c, **kwargs)
|
|
e_t_uncond = self.model.apply_model(x, t, unconditional_conditioning, **kwargs)
|
|
elif isinstance(c, dict):
|
|
e_t = self.model.apply_model(x, t, c, **kwargs)
|
|
e_t_uncond = self.model.apply_model(x, t, unconditional_conditioning, **kwargs)
|
|
else:
|
|
raise NotImplementedError
|
|
# text cfg
|
|
if uc_type is None:
|
|
e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
|
|
else:
|
|
if uc_type == 'cfg_original':
|
|
e_t = e_t + unconditional_guidance_scale * (e_t - e_t_uncond)
|
|
elif uc_type == 'cfg_ours':
|
|
e_t = e_t + unconditional_guidance_scale * (e_t_uncond - e_t)
|
|
else:
|
|
raise NotImplementedError
|
|
|
|
if score_corrector is not None:
|
|
assert self.model.parameterization == "eps"
|
|
e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs)
|
|
|
|
alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
|
|
alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev
|
|
sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
|
|
sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
|
|
# select parameters corresponding to the currently considered timestep
|
|
|
|
if is_video:
|
|
size = (b, 1, 1, 1, 1)
|
|
else:
|
|
size = (b, 1, 1, 1)
|
|
a_t = torch.full(size, alphas[index], device=device)
|
|
a_prev = torch.full(size, alphas_prev[index], device=device)
|
|
sigma_t = torch.full(size, sigmas[index], device=device)
|
|
sqrt_one_minus_at = torch.full(size, sqrt_one_minus_alphas[index],device=device)
|
|
|
|
# current prediction for x_0
|
|
pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
|
|
# print(f't={t}, pred_x0, min={torch.min(pred_x0)}, max={torch.max(pred_x0)}',file=f)
|
|
if quantize_denoised:
|
|
pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
|
|
# direction pointing to x_t
|
|
dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
|
|
|
|
if sample_noise is None:
|
|
noise = sigma_t * noise_like(x.shape, device, repeat_noise, self.noise_gen) * temperature
|
|
if noise_dropout > 0.:
|
|
noise = torch.nn.functional.dropout(noise, p=noise_dropout)
|
|
else:
|
|
noise = sigma_t * sample_noise * temperature
|
|
|
|
x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
|
|
|
|
return x_prev, pred_x0
|