1485 lines
66 KiB
Python
1485 lines
66 KiB
Python
import os
|
|
import time
|
|
import random
|
|
import itertools
|
|
from functools import partial
|
|
from contextlib import contextmanager
|
|
|
|
import numpy as np
|
|
from tqdm import tqdm
|
|
from einops import rearrange, repeat
|
|
|
|
import torch
|
|
import torch.nn as nn
|
|
import pytorch_lightning as pl
|
|
from torchvision.utils import make_grid
|
|
from torch.optim.lr_scheduler import LambdaLR
|
|
from pytorch_lightning.utilities import rank_zero_only
|
|
from videocrafter.lvdm.models.modules.distributions import normal_kl, DiagonalGaussianDistribution
|
|
from videocrafter.lvdm.models.modules.util import make_beta_schedule, extract_into_tensor, noise_like
|
|
from videocrafter.lvdm.models.modules.lora import inject_trainable_lora
|
|
from videocrafter.lvdm.samplers.ddim import DDIMSampler
|
|
from videocrafter.lvdm.utils.common_utils import log_txt_as_img, exists, default, ismap, isimage, mean_flat, count_params, instantiate_from_config, check_istarget
|
|
|
|
|
|
def disabled_train(self, mode=True):
|
|
"""Overwrite model.train with this function to make sure train/eval mode
|
|
does not change anymore."""
|
|
return self
|
|
|
|
|
|
def uniform_on_device(r1, r2, shape, device):
|
|
return (r1 - r2) * torch.rand(*shape, device=device) + r2
|
|
|
|
|
|
def split_video_to_clips(video, clip_length, drop_left=True):
|
|
video_length = video.shape[2]
|
|
shape = video.shape
|
|
if video_length % clip_length != 0 and drop_left:
|
|
video = video[:, :, :video_length // clip_length * clip_length, :, :]
|
|
print(f'[split_video_to_clips] Drop frames from {shape} to {video.shape}')
|
|
nclips = video_length // clip_length
|
|
clips = rearrange(video, 'b c (nc cl) h w -> (b nc) c cl h w', cl=clip_length, nc=nclips)
|
|
return clips
|
|
|
|
def merge_clips_to_videos(clips, bs):
|
|
nclips = clips.shape[0] // bs
|
|
video = rearrange(clips, '(b nc) c t h w -> b c (nc t) h w', nc=nclips)
|
|
return video
|
|
|
|
class DDPM(pl.LightningModule):
|
|
# classic DDPM with Gaussian diffusion, in pixel space
|
|
def __init__(self,
|
|
unet_config,
|
|
timesteps=1000,
|
|
beta_schedule="linear",
|
|
loss_type="l2",
|
|
ckpt_path=None,
|
|
ignore_keys=[],
|
|
load_only_unet=False,
|
|
monitor="val/loss",
|
|
use_ema=True,
|
|
first_stage_key="image",
|
|
image_size=256,
|
|
video_length=None,
|
|
channels=3,
|
|
log_every_t=100,
|
|
clip_denoised=True,
|
|
linear_start=1e-4,
|
|
linear_end=2e-2,
|
|
cosine_s=8e-3,
|
|
given_betas=None,
|
|
original_elbo_weight=0.,
|
|
v_posterior=0.,
|
|
l_simple_weight=1.,
|
|
conditioning_key=None,
|
|
parameterization="eps",
|
|
scheduler_config=None,
|
|
learn_logvar=False,
|
|
logvar_init=0.,
|
|
*args, **kwargs
|
|
):
|
|
super().__init__()
|
|
assert parameterization in ["eps", "x0"], 'currently only supporting "eps" and "x0"'
|
|
self.parameterization = parameterization
|
|
print(f"{self.__class__.__name__}: Running in {self.parameterization}-prediction mode")
|
|
self.cond_stage_model = None
|
|
self.clip_denoised = clip_denoised
|
|
self.log_every_t = log_every_t
|
|
self.first_stage_key = first_stage_key
|
|
self.image_size = image_size # try conv?
|
|
|
|
if isinstance(self.image_size, int):
|
|
self.image_size = [self.image_size, self.image_size]
|
|
self.channels = channels
|
|
self.model = DiffusionWrapper(unet_config, conditioning_key)
|
|
self.conditioning_key = conditioning_key # also register conditioning_key in diffusion
|
|
|
|
self.temporal_length = video_length if video_length is not None else unet_config.params.temporal_length
|
|
count_params(self.model, verbose=True)
|
|
self.use_ema = use_ema
|
|
|
|
self.use_scheduler = scheduler_config is not None
|
|
if self.use_scheduler:
|
|
self.scheduler_config = scheduler_config
|
|
|
|
self.v_posterior = v_posterior
|
|
self.original_elbo_weight = original_elbo_weight
|
|
self.l_simple_weight = l_simple_weight
|
|
|
|
if monitor is not None:
|
|
self.monitor = monitor
|
|
if ckpt_path is not None:
|
|
self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys, only_model=load_only_unet)
|
|
|
|
self.register_schedule(given_betas=given_betas, beta_schedule=beta_schedule, timesteps=timesteps,
|
|
linear_start=linear_start, linear_end=linear_end, cosine_s=cosine_s)
|
|
|
|
self.loss_type = loss_type
|
|
|
|
self.learn_logvar = learn_logvar
|
|
self.logvar = torch.full(fill_value=logvar_init, size=(self.num_timesteps,))
|
|
if self.learn_logvar:
|
|
self.logvar = nn.Parameter(self.logvar, requires_grad=True)
|
|
|
|
def register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000,
|
|
linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
|
|
if exists(given_betas):
|
|
betas = given_betas
|
|
else:
|
|
betas = make_beta_schedule(beta_schedule, timesteps, linear_start=linear_start, linear_end=linear_end,
|
|
cosine_s=cosine_s)
|
|
alphas = 1. - betas
|
|
alphas_cumprod = np.cumprod(alphas, axis=0)
|
|
alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1])
|
|
|
|
timesteps, = betas.shape
|
|
self.num_timesteps = int(timesteps)
|
|
self.linear_start = linear_start
|
|
self.linear_end = linear_end
|
|
assert alphas_cumprod.shape[0] == self.num_timesteps, 'alphas have to be defined for each timestep'
|
|
|
|
to_torch = partial(torch.tensor, dtype=torch.float32)
|
|
|
|
self.register_buffer('betas', to_torch(betas))
|
|
self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
|
|
self.register_buffer('alphas_cumprod_prev', to_torch(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)))
|
|
self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod)))
|
|
self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod)))
|
|
self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod)))
|
|
self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod - 1)))
|
|
|
|
# calculations for posterior q(x_{t-1} | x_t, x_0)
|
|
posterior_variance = (1 - self.v_posterior) * betas * (1. - alphas_cumprod_prev) / (
|
|
1. - alphas_cumprod) + self.v_posterior * betas
|
|
# above: equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t)
|
|
self.register_buffer('posterior_variance', to_torch(posterior_variance))
|
|
# below: log calculation clipped because the posterior variance is 0 at the beginning of the diffusion chain
|
|
self.register_buffer('posterior_log_variance_clipped', to_torch(np.log(np.maximum(posterior_variance, 1e-20))))
|
|
self.register_buffer('posterior_mean_coef1', to_torch(
|
|
betas * np.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod)))
|
|
self.register_buffer('posterior_mean_coef2', to_torch(
|
|
(1. - alphas_cumprod_prev) * np.sqrt(alphas) / (1. - alphas_cumprod)))
|
|
|
|
if self.parameterization == "eps":
|
|
lvlb_weights = self.betas ** 2 / (
|
|
2 * self.posterior_variance * to_torch(alphas) * (1 - self.alphas_cumprod))
|
|
elif self.parameterization == "x0":
|
|
lvlb_weights = 0.5 * np.sqrt(torch.Tensor(alphas_cumprod)) / (2. * 1 - torch.Tensor(alphas_cumprod))
|
|
else:
|
|
raise NotImplementedError("mu not supported")
|
|
# TODO how to choose this term
|
|
lvlb_weights[0] = lvlb_weights[1]
|
|
self.register_buffer('lvlb_weights', lvlb_weights, persistent=False)
|
|
assert not torch.isnan(self.lvlb_weights).all()
|
|
|
|
@contextmanager
|
|
def ema_scope(self, context=None):
|
|
if self.use_ema:
|
|
self.model_ema.store(self.model.parameters())
|
|
self.model_ema.copy_to(self.model)
|
|
if context is not None:
|
|
print(f"{context}: Switched to EMA weights")
|
|
try:
|
|
yield None
|
|
finally:
|
|
if self.use_ema:
|
|
self.model_ema.restore(self.model.parameters())
|
|
if context is not None:
|
|
print(f"{context}: Restored training weights")
|
|
|
|
def init_from_ckpt(self, path, ignore_keys=list(), only_model=False):
|
|
sd = torch.load(path, map_location="cpu")
|
|
if "state_dict" in list(sd.keys()):
|
|
sd = sd["state_dict"]
|
|
keys = list(sd.keys())
|
|
for k in keys:
|
|
for ik in ignore_keys:
|
|
if k.startswith(ik) or (ik.startswith('**') and ik.split('**')[-1] in k):
|
|
print("Deleting key {} from state_dict.".format(k))
|
|
del sd[k]
|
|
missing, unexpected = self.load_state_dict(sd, strict=False) if not only_model else self.model.load_state_dict(
|
|
sd, strict=False)
|
|
print(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys")
|
|
if len(missing) > 0:
|
|
print(f"Missing Keys: {missing}")
|
|
if len(unexpected) > 0:
|
|
print(f"Unexpected Keys: {unexpected}")
|
|
|
|
def q_mean_variance(self, x_start, t):
|
|
"""
|
|
Get the distribution q(x_t | x_0).
|
|
:param x_start: the [N x C x ...] tensor of noiseless inputs.
|
|
:param t: the number of diffusion steps (minus 1). Here, 0 means one step.
|
|
:return: A tuple (mean, variance, log_variance), all of x_start's shape.
|
|
"""
|
|
mean = (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start)
|
|
variance = extract_into_tensor(1.0 - self.alphas_cumprod, t, x_start.shape)
|
|
log_variance = extract_into_tensor(self.log_one_minus_alphas_cumprod, t, x_start.shape)
|
|
return mean, variance, log_variance
|
|
|
|
def predict_start_from_noise(self, x_t, t, noise):
|
|
return (
|
|
extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t -
|
|
extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise
|
|
)
|
|
|
|
def q_posterior(self, x_start, x_t, t):
|
|
posterior_mean = (
|
|
extract_into_tensor(self.posterior_mean_coef1, t, x_t.shape) * x_start +
|
|
extract_into_tensor(self.posterior_mean_coef2, t, x_t.shape) * x_t
|
|
)
|
|
posterior_variance = extract_into_tensor(self.posterior_variance, t, x_t.shape)
|
|
posterior_log_variance_clipped = extract_into_tensor(self.posterior_log_variance_clipped, t, x_t.shape)
|
|
return posterior_mean, posterior_variance, posterior_log_variance_clipped
|
|
|
|
def p_mean_variance(self, x, t, clip_denoised: bool):
|
|
model_out = self.model(x, t)
|
|
if self.parameterization == "eps":
|
|
x_recon = self.predict_start_from_noise(x, t=t, noise=model_out)
|
|
elif self.parameterization == "x0":
|
|
x_recon = model_out
|
|
if clip_denoised:
|
|
x_recon.clamp_(-1., 1.)
|
|
|
|
model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
|
|
return model_mean, posterior_variance, posterior_log_variance
|
|
|
|
@torch.no_grad()
|
|
def p_sample(self, x, t, clip_denoised=True, repeat_noise=False):
|
|
b, *_, device = *x.shape, x.device
|
|
model_mean, _, model_log_variance = self.p_mean_variance(x=x, t=t, clip_denoised=clip_denoised)
|
|
noise = noise_like(x.shape, device, repeat_noise)
|
|
# no noise when t == 0
|
|
nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
|
|
return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
|
|
|
|
@torch.no_grad()
|
|
def p_sample_loop(self, shape, return_intermediates=False):
|
|
device = self.betas.device
|
|
b = shape[0]
|
|
img = torch.randn(shape, device=device)
|
|
intermediates = [img]
|
|
for i in tqdm(reversed(range(0, self.num_timesteps)), desc='Sampling t', total=self.num_timesteps):
|
|
img = self.p_sample(img, torch.full((b,), i, device=device, dtype=torch.long),
|
|
clip_denoised=self.clip_denoised)
|
|
if i % self.log_every_t == 0 or i == self.num_timesteps - 1:
|
|
intermediates.append(img)
|
|
if return_intermediates:
|
|
return img, intermediates
|
|
return img
|
|
|
|
@torch.no_grad()
|
|
def sample(self, batch_size=16, return_intermediates=False):
|
|
channels = self.channels
|
|
video_length = self.total_length
|
|
size = (batch_size, channels, video_length, *self.image_size)
|
|
return self.p_sample_loop(size,
|
|
return_intermediates=return_intermediates)
|
|
|
|
def q_sample(self, x_start, t, noise=None):
|
|
noise = default(noise, lambda: torch.randn_like(x_start))
|
|
return (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
|
|
extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise)
|
|
|
|
def get_loss(self, pred, target, mean=True, mask=None):
|
|
if self.loss_type == 'l1':
|
|
loss = (target - pred).abs()
|
|
if mean:
|
|
loss = loss.mean()
|
|
elif self.loss_type == 'l2':
|
|
if mean:
|
|
loss = torch.nn.functional.mse_loss(target, pred)
|
|
else:
|
|
loss = torch.nn.functional.mse_loss(target, pred, reduction='none')
|
|
else:
|
|
raise NotImplementedError("unknown loss type '{loss_type}'")
|
|
if mask is not None:
|
|
assert(mean is False)
|
|
assert(loss.shape[2:] == mask.shape[2:]) #thw need be the same
|
|
loss = loss * mask
|
|
return loss
|
|
|
|
def p_losses(self, x_start, t, noise=None):
|
|
noise = default(noise, lambda: torch.randn_like(x_start))
|
|
x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
|
|
model_out = self.model(x_noisy, t)
|
|
|
|
loss_dict = {}
|
|
if self.parameterization == "eps":
|
|
target = noise
|
|
elif self.parameterization == "x0":
|
|
target = x_start
|
|
else:
|
|
raise NotImplementedError(f"Paramterization {self.parameterization} not yet supported")
|
|
|
|
loss = self.get_loss(model_out, target, mean=False).mean(dim=[1, 2, 3, 4])
|
|
|
|
log_prefix = 'train' if self.training else 'val'
|
|
|
|
loss_dict.update({f'{log_prefix}/loss_simple': loss.mean()})
|
|
loss_simple = loss.mean() * self.l_simple_weight
|
|
|
|
loss_vlb = (self.lvlb_weights[t] * loss).mean()
|
|
loss_dict.update({f'{log_prefix}/loss_vlb': loss_vlb})
|
|
|
|
loss = loss_simple + self.original_elbo_weight * loss_vlb
|
|
|
|
loss_dict.update({f'{log_prefix}/loss': loss})
|
|
|
|
return loss, loss_dict
|
|
|
|
def forward(self, x, *args, **kwargs):
|
|
t = torch.randint(0, self.num_timesteps, (x.shape[0],), device=self.device).long()
|
|
return self.p_losses(x, t, *args, **kwargs)
|
|
|
|
def get_input(self, batch, k):
|
|
x = batch[k]
|
|
x = x.to(memory_format=torch.contiguous_format).float()
|
|
return x
|
|
|
|
def shared_step(self, batch):
|
|
x = self.get_input(batch, self.first_stage_key)
|
|
loss, loss_dict = self(x)
|
|
return loss, loss_dict
|
|
|
|
def training_step(self, batch, batch_idx):
|
|
loss, loss_dict = self.shared_step(batch)
|
|
|
|
self.log_dict(loss_dict, prog_bar=True,
|
|
logger=True, on_step=True, on_epoch=True)
|
|
|
|
self.log("global_step", self.global_step,
|
|
prog_bar=True, logger=True, on_step=True, on_epoch=False)
|
|
|
|
if self.use_scheduler:
|
|
lr = self.optimizers().param_groups[0]['lr']
|
|
self.log('lr_abs', lr, prog_bar=True, logger=True, on_step=True, on_epoch=False)
|
|
|
|
if self.log_time:
|
|
total_train_time = (time.time() - self.start_time) / (3600*24)
|
|
avg_step_time = (time.time() - self.start_time) / (self.global_step + 1)
|
|
left_time_2w_step = (20000-self.global_step -1) * avg_step_time / (3600*24)
|
|
left_time_5w_step = (50000-self.global_step -1) * avg_step_time / (3600*24)
|
|
with open(self.logger_path, 'w') as f:
|
|
print(f'total_train_time = {total_train_time:.1f} days \n\
|
|
total_train_step = {self.global_step + 1} steps \n\
|
|
left_time_2w_step = {left_time_2w_step:.1f} days \n\
|
|
left_time_5w_step = {left_time_5w_step:.1f} days', file=f)
|
|
return loss
|
|
|
|
@torch.no_grad()
|
|
def validation_step(self, batch, batch_idx):
|
|
# _, loss_dict_no_ema = self.shared_step_validate(batch)
|
|
# with self.ema_scope():
|
|
# _, loss_dict_ema = self.shared_step_validate(batch)
|
|
# loss_dict_ema = {key + '_ema': loss_dict_ema[key] for key in loss_dict_ema}
|
|
# self.log_dict(loss_dict_no_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True)
|
|
# self.log_dict(loss_dict_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True)
|
|
if (self.global_step) % self.val_fvd_interval == 0 and self.global_step != 0:
|
|
print(f'sample for fvd...')
|
|
self.log_images_kwargs = {
|
|
'inpaint': False,
|
|
'plot_diffusion_rows': False,
|
|
'plot_progressive_rows': False,
|
|
'ddim_steps': 50,
|
|
'unconditional_guidance_scale': 15.0,
|
|
}
|
|
torch.cuda.empty_cache()
|
|
logs = self.log_images(batch, **self.log_images_kwargs)
|
|
self.log("batch_idx", batch_idx,
|
|
prog_bar=True, on_step=True, on_epoch=False)
|
|
return {'real': logs['inputs'], 'fake': logs['samples'], 'conditioning_txt_img': logs['conditioning_txt_img']}
|
|
|
|
def get_condition_validate(self, prompt):
|
|
""" text embd
|
|
"""
|
|
if isinstance(prompt, str):
|
|
prompt = [prompt]
|
|
c = self.get_learned_conditioning(prompt)
|
|
bs = c.shape[0]
|
|
|
|
return c
|
|
|
|
def on_train_batch_end(self, *args, **kwargs):
|
|
if self.use_ema:
|
|
self.model_ema(self.model)
|
|
|
|
def training_epoch_end(self, outputs):
|
|
|
|
if (self.current_epoch == 0) or self.resume_new_epoch == 0:
|
|
self.epoch_start_time = time.time()
|
|
self.current_epoch_time = 0
|
|
self.total_time = 0
|
|
self.epoch_time_avg = 0
|
|
else:
|
|
self.current_epoch_time = time.time() - self.epoch_start_time
|
|
self.epoch_start_time = time.time()
|
|
self.total_time += self.current_epoch_time
|
|
self.epoch_time_avg = self.total_time / self.current_epoch
|
|
self.resume_new_epoch += 1
|
|
epoch_avg_loss = torch.stack([x['loss'] for x in outputs]).mean()
|
|
|
|
self.log('train/epoch/loss', epoch_avg_loss, logger=True, on_epoch=True)
|
|
self.log('train/epoch/idx', self.current_epoch, logger=True, on_epoch=True)
|
|
self.log('train/epoch/time', self.current_epoch_time, logger=True, on_epoch=True)
|
|
self.log('train/epoch/time_avg', self.epoch_time_avg, logger=True, on_epoch=True)
|
|
self.log('train/epoch/time_avg_min', self.epoch_time_avg / 60, logger=True, on_epoch=True)
|
|
|
|
def _get_rows_from_list(self, samples):
|
|
n_imgs_per_row = len(samples)
|
|
denoise_grid = rearrange(samples, 'n b c t h w -> b n c t h w')
|
|
denoise_grid = rearrange(denoise_grid, 'b n c t h w -> (b n) c t h w')
|
|
denoise_grid = rearrange(denoise_grid, 'n c t h w -> (n t) c h w')
|
|
denoise_grid = make_grid(denoise_grid, nrow=n_imgs_per_row)
|
|
return denoise_grid
|
|
|
|
@torch.no_grad()
|
|
def log_images(self, batch, N=8, n_row=2, sample=True, return_keys=None,
|
|
plot_diffusion_rows=True, plot_denoise_rows=True, **kwargs):
|
|
""" log images for DDPM """
|
|
log = dict()
|
|
x = self.get_input(batch, self.first_stage_key)
|
|
N = min(x.shape[0], N)
|
|
n_row = min(x.shape[0], n_row)
|
|
x = x.to(self.device)[:N]
|
|
log["inputs"] = x
|
|
if 'fps' in batch:
|
|
log['fps'] = batch['fps']
|
|
|
|
if plot_diffusion_rows:
|
|
# get diffusion row
|
|
diffusion_row = list()
|
|
x_start = x[:n_row]
|
|
|
|
for t in range(self.num_timesteps):
|
|
if t % self.log_every_t == 0 or t == self.num_timesteps - 1:
|
|
t = repeat(torch.tensor([t]), '1 -> b', b=n_row)
|
|
t = t.to(self.device).long()
|
|
noise = torch.randn_like(x_start)
|
|
x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
|
|
diffusion_row.append(x_noisy)
|
|
|
|
log["diffusion_row"] = self._get_rows_from_list(diffusion_row)
|
|
|
|
if sample:
|
|
# get denoise row
|
|
with self.ema_scope("Plotting"):
|
|
samples, denoise_row = self.sample(batch_size=N, return_intermediates=True)
|
|
|
|
log["samples"] = samples
|
|
if plot_denoise_rows:
|
|
log["denoise_row"] = self._get_rows_from_list(denoise_row)
|
|
|
|
if return_keys:
|
|
if np.intersect1d(list(log.keys()), return_keys).shape[0] == 0:
|
|
return log
|
|
else:
|
|
return {key: log[key] for key in return_keys}
|
|
return log
|
|
|
|
def configure_optimizers(self):
|
|
lr = self.learning_rate
|
|
params = list(self.model.parameters())
|
|
if self.learn_logvar:
|
|
params = params + [self.logvar]
|
|
opt = torch.optim.AdamW(params, lr=lr)
|
|
return opt
|
|
|
|
|
|
class LatentDiffusion(DDPM):
|
|
"""main class"""
|
|
def __init__(self,
|
|
first_stage_config,
|
|
cond_stage_config,
|
|
num_timesteps_cond=None,
|
|
cond_stage_key="image",
|
|
cond_stage_trainable=False,
|
|
concat_mode=True,
|
|
cond_stage_forward=None,
|
|
conditioning_key=None,
|
|
scale_factor=1.0,
|
|
scale_by_std=False,
|
|
encoder_type="2d",
|
|
shift_factor=0.0,
|
|
split_clips=True,
|
|
downfactor_t=None,
|
|
clip_length=None,
|
|
only_model=False,
|
|
lora_args={},
|
|
*args, **kwargs):
|
|
self.num_timesteps_cond = default(num_timesteps_cond, 1)
|
|
self.scale_by_std = scale_by_std
|
|
assert self.num_timesteps_cond <= kwargs['timesteps']
|
|
# for backwards compatibility after implementation of DiffusionWrapper
|
|
|
|
if conditioning_key is None:
|
|
conditioning_key = 'concat' if concat_mode else 'crossattn'
|
|
if cond_stage_config == '__is_unconditional__':
|
|
conditioning_key = None
|
|
ckpt_path = kwargs.pop("ckpt_path", None)
|
|
ignore_keys = kwargs.pop("ignore_keys", [])
|
|
super().__init__(conditioning_key=conditioning_key, *args, **kwargs)
|
|
self.concat_mode = concat_mode
|
|
self.cond_stage_trainable = cond_stage_trainable
|
|
self.cond_stage_key = cond_stage_key
|
|
try:
|
|
self.num_downs = len(first_stage_config.params.ddconfig.ch_mult) - 1
|
|
except:
|
|
self.num_downs = 0
|
|
if not scale_by_std:
|
|
self.scale_factor = scale_factor
|
|
else:
|
|
self.register_buffer('scale_factor', torch.tensor(scale_factor))
|
|
self.instantiate_first_stage(first_stage_config)
|
|
self.instantiate_cond_stage(cond_stage_config)
|
|
self.cond_stage_forward = cond_stage_forward
|
|
self.clip_denoised = False
|
|
self.bbox_tokenizer = None
|
|
self.cond_stage_config = cond_stage_config
|
|
self.first_stage_config = first_stage_config
|
|
self.encoder_type = encoder_type
|
|
assert(encoder_type in ["2d", "3d"])
|
|
self.restarted_from_ckpt = False
|
|
self.shift_factor = shift_factor
|
|
if ckpt_path is not None:
|
|
self.init_from_ckpt(ckpt_path, ignore_keys, only_model=only_model)
|
|
self.restarted_from_ckpt = True
|
|
self.split_clips = split_clips
|
|
self.downfactor_t = downfactor_t
|
|
self.clip_length = clip_length
|
|
# lora related args
|
|
self.inject_unet = getattr(lora_args, "inject_unet", False)
|
|
self.inject_clip = getattr(lora_args, "inject_clip", False)
|
|
self.inject_unet_key_word = getattr(lora_args, "inject_unet_key_word", None)
|
|
self.inject_clip_key_word = getattr(lora_args, "inject_clip_key_word", None)
|
|
self.lora_rank = getattr(lora_args, "lora_rank", 4)
|
|
|
|
def make_cond_schedule(self, ):
|
|
self.cond_ids = torch.full(size=(self.num_timesteps,), fill_value=self.num_timesteps - 1, dtype=torch.long)
|
|
ids = torch.round(torch.linspace(0, self.num_timesteps - 1, self.num_timesteps_cond)).long()
|
|
self.cond_ids[:self.num_timesteps_cond] = ids
|
|
|
|
def inject_lora(self, lora_scale=1.0):
|
|
if self.inject_unet:
|
|
self.lora_require_grad_params, self.lora_names = inject_trainable_lora(self.model, self.inject_unet_key_word,
|
|
r=self.lora_rank,
|
|
scale=lora_scale
|
|
)
|
|
if self.inject_clip:
|
|
self.lora_require_grad_params_clip, self.lora_names_clip = inject_trainable_lora(self.cond_stage_model, self.inject_clip_key_word,
|
|
r=self.lora_rank,
|
|
scale=lora_scale
|
|
)
|
|
|
|
@rank_zero_only
|
|
@torch.no_grad()
|
|
def on_train_batch_start(self, batch, batch_idx, dataloader_idx=None):
|
|
# only for very first batch, reset the self.scale_factor
|
|
if self.scale_by_std and self.current_epoch == 0 and self.global_step == 0 and batch_idx == 0 and not self.restarted_from_ckpt:
|
|
assert self.scale_factor == 1., 'rather not use custom rescaling and std-rescaling simultaneously'
|
|
# set rescale weight to 1./std of encodings
|
|
print("### USING STD-RESCALING ###")
|
|
x = super().get_input(batch, self.first_stage_key)
|
|
x = x.to(self.device)
|
|
encoder_posterior = self.encode_first_stage(x)
|
|
z = self.get_first_stage_encoding(encoder_posterior).detach()
|
|
del self.scale_factor
|
|
self.register_buffer('scale_factor', 1. / z.flatten().std())
|
|
print(f"setting self.scale_factor to {self.scale_factor}")
|
|
print("### USING STD-RESCALING ###")
|
|
print(f"std={z.flatten().std()}")
|
|
|
|
def register_schedule(self,
|
|
given_betas=None, beta_schedule="linear", timesteps=1000,
|
|
linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
|
|
super().register_schedule(given_betas, beta_schedule, timesteps, linear_start, linear_end, cosine_s)
|
|
|
|
self.shorten_cond_schedule = self.num_timesteps_cond > 1
|
|
if self.shorten_cond_schedule:
|
|
self.make_cond_schedule()
|
|
|
|
def instantiate_first_stage(self, config):
|
|
model = instantiate_from_config(config)
|
|
self.first_stage_model = model.eval()
|
|
self.first_stage_model.train = disabled_train
|
|
for param in self.first_stage_model.parameters():
|
|
param.requires_grad = False
|
|
|
|
def instantiate_cond_stage(self, config):
|
|
if config is None:
|
|
self.cond_stage_model = None
|
|
return
|
|
if not self.cond_stage_trainable:
|
|
if config == "__is_first_stage__":
|
|
print("Using first stage also as cond stage.")
|
|
self.cond_stage_model = self.first_stage_model
|
|
elif config == "__is_unconditional__":
|
|
print(f"Training {self.__class__.__name__} as an unconditional model.")
|
|
self.cond_stage_model = None
|
|
else:
|
|
model = instantiate_from_config(config)
|
|
self.cond_stage_model = model.eval()
|
|
self.cond_stage_model.train = disabled_train
|
|
for param in self.cond_stage_model.parameters():
|
|
param.requires_grad = False
|
|
else:
|
|
assert config != '__is_first_stage__'
|
|
assert config != '__is_unconditional__'
|
|
model = instantiate_from_config(config)
|
|
self.cond_stage_model = model
|
|
|
|
|
|
def get_first_stage_encoding(self, encoder_posterior, noise=None):
|
|
if isinstance(encoder_posterior, DiagonalGaussianDistribution):
|
|
z = encoder_posterior.sample(noise=noise)
|
|
elif isinstance(encoder_posterior, torch.Tensor):
|
|
z = encoder_posterior
|
|
else:
|
|
raise NotImplementedError(f"encoder_posterior of type '{type(encoder_posterior)}' not yet implemented")
|
|
z = self.scale_factor * (z + self.shift_factor)
|
|
return z
|
|
|
|
|
|
def get_learned_conditioning(self, c):
|
|
if self.cond_stage_forward is None:
|
|
if hasattr(self.cond_stage_model, 'encode') and callable(self.cond_stage_model.encode):
|
|
c = self.cond_stage_model.encode(c)
|
|
if isinstance(c, DiagonalGaussianDistribution):
|
|
c = c.mode()
|
|
else:
|
|
c = self.cond_stage_model(c)
|
|
else:
|
|
assert hasattr(self.cond_stage_model, self.cond_stage_forward)
|
|
c = getattr(self.cond_stage_model, self.cond_stage_forward)(c)
|
|
return c
|
|
|
|
|
|
@torch.no_grad()
|
|
def get_condition(self, batch, x, bs, force_c_encode, k, cond_key, is_imgs=False):
|
|
is_conditional = self.model.conditioning_key is not None # crossattn
|
|
if is_conditional:
|
|
if cond_key is None:
|
|
cond_key = self.cond_stage_key
|
|
|
|
# get condition batch of different condition type
|
|
if cond_key != self.first_stage_key:
|
|
assert(cond_key in ["caption", "txt"])
|
|
xc = batch[cond_key]
|
|
else:
|
|
xc = x
|
|
|
|
# if static video
|
|
if self.static_video:
|
|
xc_ = [c + ' (static)' for c in xc]
|
|
xc = xc_
|
|
|
|
# get learned condition.
|
|
# can directly skip it: c = xc
|
|
if self.cond_stage_config is not None and (not self.cond_stage_trainable or force_c_encode):
|
|
if isinstance(xc, torch.Tensor):
|
|
xc = xc.to(self.device)
|
|
c = self.get_learned_conditioning(xc)
|
|
else:
|
|
c = xc
|
|
|
|
if self.classfier_free_guidance:
|
|
if cond_key in ['caption', "txt"] and self.uncond_type == 'empty_seq':
|
|
for i, ci in enumerate(c):
|
|
if random.random() < self.prob:
|
|
c[i] = ""
|
|
elif cond_key == 'class_label' and self.uncond_type == 'zero_embed':
|
|
pass
|
|
elif cond_key == 'class_label' and self.uncond_type == 'learned_embed':
|
|
import pdb;pdb.set_trace()
|
|
for i, ci in enumerate(c):
|
|
if random.random() < self.prob:
|
|
c[i]['class_label'] = self.n_classes
|
|
|
|
else:
|
|
raise NotImplementedError
|
|
|
|
if self.zero_cond_embed:
|
|
import pdb;pdb.set_trace()
|
|
c = torch.zeros_like(c)
|
|
|
|
# process c
|
|
if bs is not None:
|
|
if (is_imgs and not self.static_video):
|
|
c = c[:bs*self.temporal_length] # each random img (in T axis) has a corresponding prompt
|
|
else:
|
|
c = c[:bs]
|
|
|
|
else:
|
|
c = None
|
|
xc = None
|
|
|
|
return c, xc
|
|
|
|
@torch.no_grad()
|
|
def get_input(self, batch, k, return_first_stage_outputs=False, force_c_encode=False,
|
|
cond_key=None, return_original_cond=False, bs=None, mask_temporal=False):
|
|
""" Get input in LDM
|
|
"""
|
|
# get input imgaes
|
|
x = super().get_input(batch, k) # k = first_stage_key=image
|
|
is_imgs = True if k == 'jpg' else False
|
|
if is_imgs:
|
|
if self.static_video:
|
|
# repeat single img to a static video
|
|
x = x.unsqueeze(2) # bchw -> bc1hw
|
|
x = x.repeat(1,1,self.temporal_length,1,1) # bc1hw -> bcthw
|
|
else:
|
|
# rearrange to videos with T random img
|
|
bs_load = x.shape[0] // self.temporal_length
|
|
x = x[:bs_load*self.temporal_length, ...]
|
|
x = rearrange(x, '(b t) c h w -> b c t h w', t=self.temporal_length, b=bs_load)
|
|
|
|
if bs is not None:
|
|
x = x[:bs]
|
|
|
|
x = x.to(self.device)
|
|
x_ori = x
|
|
|
|
b, _, t, h, w = x.shape
|
|
|
|
# encode video frames x to z via a 2D encoder
|
|
x = rearrange(x, 'b c t h w -> (b t) c h w')
|
|
encoder_posterior = self.encode_first_stage(x, mask_temporal)
|
|
z = self.get_first_stage_encoding(encoder_posterior).detach()
|
|
z = rearrange(z, '(b t) c h w -> b c t h w', b=b, t=t)
|
|
|
|
|
|
c, xc = self.get_condition(batch, x, bs, force_c_encode, k, cond_key, is_imgs)
|
|
out = [z, c]
|
|
|
|
if return_first_stage_outputs:
|
|
xrec = self.decode_first_stage(z, mask_temporal=mask_temporal)
|
|
out.extend([x_ori, xrec])
|
|
if return_original_cond:
|
|
if isinstance(xc, torch.Tensor) and xc.dim() == 4:
|
|
xc = rearrange(xc, '(b t) c h w -> b c t h w', b=b, t=t)
|
|
out.append(xc)
|
|
|
|
return out
|
|
|
|
@torch.no_grad()
|
|
def decode(self, z, **kwargs,):
|
|
z = 1. / self.scale_factor * z - self.shift_factor
|
|
results = self.first_stage_model.decode(z,**kwargs)
|
|
return results
|
|
|
|
@torch.no_grad()
|
|
def decode_first_stage_2DAE(self, z, decode_bs=16, return_cpu=True, **kwargs):
|
|
b, _, t, _, _ = z.shape
|
|
z = rearrange(z, 'b c t h w -> (b t) c h w')
|
|
if decode_bs is None:
|
|
results = self.decode(z, **kwargs)
|
|
else:
|
|
z = torch.split(z, decode_bs, dim=0)
|
|
if return_cpu:
|
|
results = torch.cat([self.decode(z_, **kwargs).cpu() for z_ in z], dim=0)
|
|
else:
|
|
results = torch.cat([self.decode(z_, **kwargs) for z_ in z], dim=0)
|
|
results = rearrange(results, '(b t) c h w -> b c t h w', b=b,t=t).contiguous()
|
|
return results
|
|
|
|
@torch.no_grad()
|
|
def decode_first_stage(self, z, decode_bs=16, return_cpu=True, **kwargs):
|
|
assert(self.encoder_type == "2d" and z.dim() == 5)
|
|
return self.decode_first_stage_2DAE(z, decode_bs=decode_bs, return_cpu=return_cpu, **kwargs)
|
|
|
|
@torch.no_grad()
|
|
def encode_first_stage_2DAE(self, x, encode_bs=16):
|
|
b, _, t, _, _ = x.shape
|
|
x = rearrange(x, 'b c t h w -> (b t) c h w')
|
|
if encode_bs is None:
|
|
results = self.first_stage_model.encode(x)
|
|
else:
|
|
x = torch.split(x, encode_bs, dim=0)
|
|
zs = []
|
|
for x_ in x:
|
|
encoder_posterior = self.first_stage_model.encode(x_)
|
|
z = self.get_first_stage_encoding(encoder_posterior).detach()
|
|
zs.append(z)
|
|
results = torch.cat(zs, dim=0)
|
|
results = rearrange(results, '(b t) c h w -> b c t h w', b=b,t=t)
|
|
return results
|
|
|
|
@torch.no_grad()
|
|
def encode_first_stage(self, x):
|
|
assert(self.encoder_type == "2d" and x.dim() == 5)
|
|
b, _, t, _, _ = x.shape
|
|
x = rearrange(x, 'b c t h w -> (b t) c h w')
|
|
results = self.first_stage_model.encode(x)
|
|
results = rearrange(results, '(b t) c h w -> b c t h w', b=b,t=t)
|
|
return results
|
|
|
|
def shared_step(self, batch, **kwargs):
|
|
""" shared step of LDM.
|
|
If learned condition, c is raw condition (e.g. text)
|
|
Encoding condition is performed in below forward function.
|
|
"""
|
|
x, c = self.get_input(batch, self.first_stage_key)
|
|
loss = self(x, c)
|
|
return loss
|
|
|
|
def forward(self, x, c, *args, **kwargs):
|
|
start_t = getattr(self, "start_t", 0)
|
|
end_t = getattr(self, "end_t", self.num_timesteps)
|
|
t = torch.randint(start_t, end_t, (x.shape[0],), device=self.device).long()
|
|
|
|
if self.model.conditioning_key is not None:
|
|
assert c is not None
|
|
if self.cond_stage_trainable:
|
|
c = self.get_learned_conditioning(c)
|
|
if self.classfier_free_guidance and self.uncond_type == 'zero_embed':
|
|
for i, ci in enumerate(c):
|
|
if random.random() < self.prob:
|
|
c[i] = torch.zeros_like(c[i])
|
|
if self.shorten_cond_schedule: # TODO: drop this option
|
|
tc = self.cond_ids[t].to(self.device)
|
|
c = self.q_sample(x_start=c, t=tc, noise=torch.randn_like(c.float()))
|
|
|
|
return self.p_losses(x, c, t, *args, **kwargs)
|
|
|
|
def apply_model(self, x_noisy, t, cond, return_ids=False, **kwargs):
|
|
|
|
if isinstance(cond, dict):
|
|
# hybrid case, cond is exptected to be a dict
|
|
pass
|
|
else:
|
|
if not isinstance(cond, list):
|
|
cond = [cond]
|
|
key = 'c_concat' if self.model.conditioning_key == 'concat' else 'c_crossattn'
|
|
cond = {key: cond}
|
|
|
|
x_recon = self.model(x_noisy, t, **cond, **kwargs)
|
|
|
|
if isinstance(x_recon, tuple) and not return_ids:
|
|
return x_recon[0]
|
|
else:
|
|
return x_recon
|
|
|
|
def _predict_eps_from_xstart(self, x_t, t, pred_xstart):
|
|
return (extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t - pred_xstart) / \
|
|
extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape)
|
|
|
|
def _prior_bpd(self, x_start):
|
|
"""
|
|
Get the prior KL term for the variational lower-bound, measured in
|
|
bits-per-dim.
|
|
This term can't be optimized, as it only depends on the encoder.
|
|
:param x_start: the [N x C x ...] tensor of inputs.
|
|
:return: a batch of [N] KL values (in bits), one per batch element.
|
|
"""
|
|
batch_size = x_start.shape[0]
|
|
t = torch.tensor([self.num_timesteps - 1] * batch_size, device=x_start.device)
|
|
qt_mean, _, qt_log_variance = self.q_mean_variance(x_start, t)
|
|
kl_prior = normal_kl(mean1=qt_mean, logvar1=qt_log_variance, mean2=0.0, logvar2=0.0)
|
|
return mean_flat(kl_prior) / np.log(2.0)
|
|
|
|
def p_losses(self, x_start, cond, t, noise=None, skip_qsample=False, x_noisy=None, cond_mask=None, **kwargs,):
|
|
if not skip_qsample:
|
|
noise = default(noise, lambda: torch.randn_like(x_start))
|
|
x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise)
|
|
else:
|
|
assert(x_noisy is not None)
|
|
assert(noise is not None)
|
|
model_output = self.apply_model(x_noisy, t, cond, **kwargs)
|
|
|
|
loss_dict = {}
|
|
prefix = 'train' if self.training else 'val'
|
|
|
|
if self.parameterization == "x0":
|
|
target = x_start
|
|
elif self.parameterization == "eps":
|
|
target = noise
|
|
else:
|
|
raise NotImplementedError()
|
|
|
|
loss_simple = self.get_loss(model_output, target, mean=False).mean([1, 2, 3, 4])
|
|
loss_dict.update({f'{prefix}/loss_simple': loss_simple.mean()})
|
|
if self.logvar.device != self.device:
|
|
self.logvar = self.logvar.to(self.device)
|
|
logvar_t = self.logvar[t]
|
|
loss = loss_simple / torch.exp(logvar_t) + logvar_t
|
|
if self.learn_logvar:
|
|
loss_dict.update({f'{prefix}/loss_gamma': loss.mean()})
|
|
loss_dict.update({'logvar': self.logvar.data.mean()})
|
|
|
|
loss = self.l_simple_weight * loss.mean()
|
|
|
|
loss_vlb = self.get_loss(model_output, target, mean=False).mean(dim=(1, 2, 3, 4))
|
|
loss_vlb = (self.lvlb_weights[t] * loss_vlb).mean()
|
|
loss_dict.update({f'{prefix}/loss_vlb': loss_vlb})
|
|
loss += (self.original_elbo_weight * loss_vlb)
|
|
loss_dict.update({f'{prefix}/loss': loss})
|
|
|
|
return loss, loss_dict
|
|
|
|
def p_mean_variance(self, x, c, t, clip_denoised: bool, return_codebook_ids=False, quantize_denoised=False,
|
|
return_x0=False, score_corrector=None, corrector_kwargs=None,
|
|
unconditional_guidance_scale=1., unconditional_conditioning=None,
|
|
uc_type=None,):
|
|
t_in = t
|
|
if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
|
|
model_out = self.apply_model(x, t_in, c, return_ids=return_codebook_ids)
|
|
else:
|
|
# with unconditional condition
|
|
if isinstance(c, torch.Tensor):
|
|
x_in = torch.cat([x] * 2)
|
|
t_in = torch.cat([t] * 2)
|
|
c_in = torch.cat([unconditional_conditioning, c])
|
|
model_out_uncond, model_out = self.apply_model(x_in, t_in, c_in, return_ids=return_codebook_ids).chunk(2)
|
|
elif isinstance(c, dict):
|
|
model_out = self.apply_model(x, t, c, return_ids=return_codebook_ids)
|
|
model_out_uncond = self.apply_model(x, t, unconditional_conditioning, return_ids=return_codebook_ids)
|
|
else:
|
|
raise NotImplementedError
|
|
if uc_type is None:
|
|
model_out = model_out_uncond + unconditional_guidance_scale * (model_out - model_out_uncond)
|
|
else:
|
|
if uc_type == 'cfg_original':
|
|
model_out = model_out + unconditional_guidance_scale * (model_out - model_out_uncond)
|
|
elif uc_type == 'cfg_ours':
|
|
model_out = model_out + unconditional_guidance_scale * (model_out_uncond - model_out)
|
|
else:
|
|
raise NotImplementedError
|
|
|
|
if score_corrector is not None:
|
|
assert self.parameterization == "eps"
|
|
model_out = score_corrector.modify_score(self, model_out, x, t, c, **corrector_kwargs)
|
|
|
|
if return_codebook_ids:
|
|
model_out, logits = model_out
|
|
|
|
if self.parameterization == "eps":
|
|
x_recon = self.predict_start_from_noise(x, t=t, noise=model_out)
|
|
elif self.parameterization == "x0":
|
|
x_recon = model_out
|
|
else:
|
|
raise NotImplementedError()
|
|
|
|
if clip_denoised:
|
|
x_recon.clamp_(-1., 1.)
|
|
if quantize_denoised:
|
|
x_recon, _, [_, _, indices] = self.first_stage_model.quantize(x_recon)
|
|
model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t)
|
|
if return_codebook_ids:
|
|
return model_mean, posterior_variance, posterior_log_variance, logits
|
|
elif return_x0:
|
|
return model_mean, posterior_variance, posterior_log_variance, x_recon
|
|
else:
|
|
return model_mean, posterior_variance, posterior_log_variance
|
|
|
|
@torch.no_grad()
|
|
def p_sample(self, x, c, t, clip_denoised=False, repeat_noise=False,
|
|
return_codebook_ids=False, quantize_denoised=False, return_x0=False,
|
|
temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
|
|
unconditional_guidance_scale=1., unconditional_conditioning=None,
|
|
uc_type=None,):
|
|
b, *_, device = *x.shape, x.device
|
|
outputs = self.p_mean_variance(x=x, c=c, t=t, clip_denoised=clip_denoised,
|
|
return_codebook_ids=return_codebook_ids,
|
|
quantize_denoised=quantize_denoised,
|
|
return_x0=return_x0,
|
|
score_corrector=score_corrector, corrector_kwargs=corrector_kwargs,
|
|
unconditional_guidance_scale=unconditional_guidance_scale,
|
|
unconditional_conditioning=unconditional_conditioning,
|
|
uc_type=uc_type,)
|
|
if return_codebook_ids:
|
|
raise DeprecationWarning("Support dropped.")
|
|
elif return_x0:
|
|
model_mean, _, model_log_variance, x0 = outputs
|
|
else:
|
|
model_mean, _, model_log_variance = outputs
|
|
|
|
noise = noise_like(x.shape, device, repeat_noise) * temperature
|
|
if noise_dropout > 0.:
|
|
noise = torch.nn.functional.dropout(noise, p=noise_dropout)
|
|
|
|
nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1)))
|
|
|
|
if return_codebook_ids:
|
|
return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise, logits.argmax(dim=1)
|
|
if return_x0:
|
|
return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise, x0
|
|
else:
|
|
return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise
|
|
|
|
@torch.no_grad()
|
|
def progressive_denoising(self, cond, shape, verbose=True, callback=None, quantize_denoised=False,
|
|
img_callback=None, mask=None, x0=None, temperature=1., noise_dropout=0.,
|
|
score_corrector=None, corrector_kwargs=None, batch_size=None, x_T=None, start_T=None,
|
|
log_every_t=None):
|
|
if not log_every_t:
|
|
log_every_t = self.log_every_t
|
|
timesteps = self.num_timesteps
|
|
if batch_size is not None:
|
|
b = batch_size if batch_size is not None else shape[0]
|
|
shape = [batch_size] + list(shape)
|
|
else:
|
|
b = batch_size = shape[0]
|
|
if x_T is None:
|
|
img = torch.randn(shape, device=self.device)
|
|
else:
|
|
img = x_T
|
|
intermediates = []
|
|
if cond is not None:
|
|
if isinstance(cond, dict):
|
|
cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else
|
|
list(map(lambda x: x[:batch_size], cond[key])) for key in cond}
|
|
else:
|
|
cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size]
|
|
|
|
if start_T is not None:
|
|
timesteps = min(timesteps, start_T)
|
|
iterator = tqdm(reversed(range(0, timesteps)), desc='Progressive Generation',
|
|
total=timesteps) if verbose else reversed(
|
|
range(0, timesteps))
|
|
if type(temperature) == float:
|
|
temperature = [temperature] * timesteps
|
|
|
|
for i in iterator:
|
|
ts = torch.full((b,), i, device=self.device, dtype=torch.long)
|
|
if self.shorten_cond_schedule:
|
|
assert self.model.conditioning_key != 'hybrid'
|
|
tc = self.cond_ids[ts].to(cond.device)
|
|
cond = self.q_sample(x_start=cond, t=tc, noise=torch.randn_like(cond))
|
|
|
|
img, x0_partial = self.p_sample(img, cond, ts,
|
|
clip_denoised=self.clip_denoised,
|
|
quantize_denoised=quantize_denoised, return_x0=True,
|
|
temperature=temperature[i], noise_dropout=noise_dropout,
|
|
score_corrector=score_corrector, corrector_kwargs=corrector_kwargs)
|
|
if mask is not None:
|
|
assert x0 is not None
|
|
img_orig = self.q_sample(x0, ts)
|
|
img = img_orig * mask + (1. - mask) * img
|
|
|
|
if i % log_every_t == 0 or i == timesteps - 1:
|
|
intermediates.append(x0_partial)
|
|
if callback: callback(i)
|
|
if img_callback: img_callback(img, i)
|
|
return img, intermediates
|
|
|
|
@torch.no_grad()
|
|
def p_sample_loop(self, cond, shape, return_intermediates=False,
|
|
x_T=None, verbose=True, callback=None, timesteps=None, quantize_denoised=False,
|
|
mask=None, x0=None, img_callback=None, start_T=None,
|
|
log_every_t=None,
|
|
unconditional_guidance_scale=1., unconditional_conditioning=None,
|
|
uc_type=None,):
|
|
|
|
if not log_every_t:
|
|
log_every_t = self.log_every_t
|
|
device = self.betas.device
|
|
b = shape[0]
|
|
|
|
# sample an initial noise
|
|
if x_T is None:
|
|
img = torch.randn(shape, device=device)
|
|
else:
|
|
img = x_T
|
|
|
|
intermediates = [img]
|
|
if timesteps is None:
|
|
timesteps = self.num_timesteps
|
|
|
|
if start_T is not None:
|
|
timesteps = min(timesteps, start_T)
|
|
iterator = tqdm(reversed(range(0, timesteps)), desc='Sampling t', total=timesteps) if verbose else reversed(
|
|
range(0, timesteps))
|
|
|
|
if mask is not None:
|
|
assert x0 is not None
|
|
assert x0.shape[2:3] == mask.shape[2:3] # spatial size has to match
|
|
|
|
for i in iterator:
|
|
ts = torch.full((b,), i, device=device, dtype=torch.long)
|
|
if self.shorten_cond_schedule:
|
|
assert self.model.conditioning_key != 'hybrid'
|
|
tc = self.cond_ids[ts].to(cond.device)
|
|
cond = self.q_sample(x_start=cond, t=tc, noise=torch.randn_like(cond))
|
|
|
|
img = self.p_sample(img, cond, ts,
|
|
clip_denoised=self.clip_denoised,
|
|
quantize_denoised=quantize_denoised,
|
|
unconditional_guidance_scale=unconditional_guidance_scale,
|
|
unconditional_conditioning=unconditional_conditioning,
|
|
uc_type=uc_type)
|
|
if mask is not None:
|
|
img_orig = self.q_sample(x0, ts)
|
|
img = img_orig * mask + (1. - mask) * img
|
|
|
|
if i % log_every_t == 0 or i == timesteps - 1:
|
|
intermediates.append(img)
|
|
if callback: callback(i)
|
|
if img_callback: img_callback(img, i)
|
|
|
|
if return_intermediates:
|
|
return img, intermediates
|
|
return img
|
|
|
|
@torch.no_grad()
|
|
def sample(self, cond, batch_size=16, return_intermediates=False, x_T=None,
|
|
verbose=True, timesteps=None, quantize_denoised=False,
|
|
mask=None, x0=None, shape=None, **kwargs):
|
|
if shape is None:
|
|
shape = (batch_size, self.channels, self.total_length, *self.image_size)
|
|
if cond is not None:
|
|
if isinstance(cond, dict):
|
|
cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else
|
|
list(map(lambda x: x[:batch_size], cond[key])) for key in cond}
|
|
else:
|
|
cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size]
|
|
return self.p_sample_loop(cond,
|
|
shape,
|
|
return_intermediates=return_intermediates, x_T=x_T,
|
|
verbose=verbose, timesteps=timesteps, quantize_denoised=quantize_denoised,
|
|
mask=mask, x0=x0,)
|
|
|
|
@torch.no_grad()
|
|
def sample_log(self,cond,batch_size,ddim, ddim_steps,**kwargs):
|
|
|
|
if ddim:
|
|
ddim_sampler = DDIMSampler(self)
|
|
shape = (self.channels, self.total_length, *self.image_size)
|
|
samples, intermediates =ddim_sampler.sample(ddim_steps,batch_size,
|
|
shape,cond,verbose=False, **kwargs)
|
|
|
|
else:
|
|
samples, intermediates = self.sample(cond=cond, batch_size=batch_size,
|
|
return_intermediates=True, **kwargs)
|
|
|
|
return samples, intermediates
|
|
|
|
@torch.no_grad()
|
|
def log_condition(self, log, batch, xc, x, c, cond_stage_key=None):
|
|
"""
|
|
xc: oringinal condition before enconding.
|
|
c: condition after encoding.
|
|
"""
|
|
if x.dim() == 5:
|
|
txt_img_shape = [x.shape[3], x.shape[4]]
|
|
elif x.dim() == 4:
|
|
txt_img_shape = [x.shape[2], x.shape[3]]
|
|
else:
|
|
raise ValueError
|
|
if self.model.conditioning_key is not None: #concat-time-mask
|
|
if hasattr(self.cond_stage_model, "decode"):
|
|
xc = self.cond_stage_model.decode(c)
|
|
log["conditioning"] = xc
|
|
elif cond_stage_key in ["caption", "txt"]:
|
|
log["conditioning_txt_img"] = log_txt_as_img(txt_img_shape, batch[cond_stage_key], size=x.shape[3]//25)
|
|
log["conditioning_txt"] = batch[cond_stage_key]
|
|
elif cond_stage_key == 'class_label':
|
|
try:
|
|
xc = log_txt_as_img(txt_img_shape, batch["human_label"], size=x.shape[3]//25)
|
|
except:
|
|
xc = log_txt_as_img(txt_img_shape, batch["class_name"], size=x.shape[3]//25)
|
|
log['conditioning'] = xc
|
|
elif isimage(xc):
|
|
log["conditioning"] = xc
|
|
if ismap(xc):
|
|
log["original_conditioning"] = self.to_rgb(xc)
|
|
if isinstance(c, dict) and 'mask' in c:
|
|
log['mask'] =self.mask_to_rgb(c['mask'])
|
|
return log
|
|
|
|
@torch.no_grad()
|
|
def log_images(self, batch, N=8, n_row=4, sample=True, ddim_steps=200, ddim_eta=1., unconditional_guidance_scale=1.0,
|
|
first_stage_key2=None, cond_key2=None,
|
|
c=None,
|
|
**kwargs):
|
|
""" log images for LatentDiffusion """
|
|
use_ddim = ddim_steps is not None
|
|
is_imgs = first_stage_key2 is not None
|
|
if is_imgs:
|
|
assert(cond_key2 is not None)
|
|
log = dict()
|
|
|
|
# get input
|
|
z, c, x, xrec, xc = self.get_input(batch,
|
|
k=self.first_stage_key if first_stage_key2 is None else first_stage_key2,
|
|
return_first_stage_outputs=True,
|
|
force_c_encode=True,
|
|
return_original_cond=True,
|
|
bs=N,
|
|
cond_key=cond_key2 if cond_key2 is not None else None,
|
|
)
|
|
|
|
N_ori = N
|
|
N = min(z.shape[0], N)
|
|
n_row = min(x.shape[0], n_row)
|
|
|
|
if unconditional_guidance_scale != 1.0:
|
|
prompts = N * self.temporal_length * [""] if (is_imgs and not self.static_video) else N * [""]
|
|
uc = self.get_condition_validate(prompts)
|
|
|
|
else:
|
|
uc = None
|
|
|
|
log["inputs"] = x
|
|
log["reconstruction"] = xrec
|
|
log = self.log_condition(log, batch, xc, x, c,
|
|
cond_stage_key=self.cond_stage_key if cond_key2 is None else cond_key2
|
|
)
|
|
|
|
if sample:
|
|
with self.ema_scope("Plotting"):
|
|
samples, z_denoise_row = self.sample_log(cond=c,batch_size=N,ddim=use_ddim,
|
|
ddim_steps=ddim_steps,eta=ddim_eta,
|
|
temporal_length=self.video_length,
|
|
unconditional_guidance_scale=unconditional_guidance_scale,
|
|
unconditional_conditioning=uc, **kwargs,
|
|
)
|
|
# decode samples
|
|
x_samples = self.decode_first_stage(samples)
|
|
log["samples"] = x_samples
|
|
return log
|
|
|
|
def configure_optimizers(self):
|
|
""" configure_optimizers for LatentDiffusion """
|
|
lr = self.learning_rate
|
|
|
|
# --------------------------------------------------------------------------------
|
|
# set parameters
|
|
if hasattr(self, "only_optimize_empty_parameters") and self.only_optimize_empty_parameters:
|
|
print("[INFO] Optimize only empty parameters!")
|
|
assert(hasattr(self, "empty_paras"))
|
|
params = [p for n, p in self.model.named_parameters() if n in self.empty_paras]
|
|
elif hasattr(self, "only_optimize_pretrained_parameters") and self.only_optimize_pretrained_parameters:
|
|
print("[INFO] Optimize only pretrained parameters!")
|
|
assert(hasattr(self, "empty_paras"))
|
|
params = [p for n, p in self.model.named_parameters() if n not in self.empty_paras]
|
|
assert(len(params) != 0)
|
|
elif getattr(self, "optimize_empty_and_spatialattn", False):
|
|
print("[INFO] Optimize empty parameters + spatial transformer!")
|
|
assert(hasattr(self, "empty_paras"))
|
|
empty_paras = [p for n, p in self.model.named_parameters() if n in self.empty_paras]
|
|
SA_list = [".attn1.", ".attn2.", ".ff.", ".norm1.", ".norm2.", ".norm3."]
|
|
SA_params = [p for n, p in self.model.named_parameters() if check_istarget(n, SA_list)]
|
|
if getattr(self, "spatial_lr_decay", False):
|
|
params = [
|
|
{"params": empty_paras},
|
|
{"params": SA_params, "lr": lr * self.spatial_lr_decay}
|
|
]
|
|
else:
|
|
params = empty_paras + SA_params
|
|
else:
|
|
# optimize whole denoiser
|
|
if hasattr(self, "spatial_lr_decay") and self.spatial_lr_decay:
|
|
print("[INFO] Optimize the whole net with different lr!")
|
|
print(f"[INFO] {lr} for empty paras, {lr * self.spatial_lr_decay} for pretrained paras!")
|
|
empty_paras = [p for n, p in self.model.named_parameters() if n in self.empty_paras]
|
|
# assert(len(empty_paras) == len(self.empty_paras)) # self.empty_paras:cond_stage_model.embedding.weight not in diffusion model params
|
|
pretrained_paras = [p for n, p in self.model.named_parameters() if n not in self.empty_paras]
|
|
params = [
|
|
{"params": empty_paras},
|
|
{"params": pretrained_paras, "lr": lr * self.spatial_lr_decay}
|
|
]
|
|
print(f"[INFO] Empty paras: {len(empty_paras)}, Pretrained paras: {len(pretrained_paras)}")
|
|
|
|
else:
|
|
params = list(self.model.parameters())
|
|
|
|
if hasattr(self, "generator_trainable") and not self.generator_trainable:
|
|
# fix unet denoiser
|
|
params = list()
|
|
|
|
if self.inject_unet:
|
|
params = itertools.chain(*self.lora_require_grad_params)
|
|
|
|
if self.inject_clip:
|
|
if self.inject_unet:
|
|
params = list(params)+list(itertools.chain(*self.lora_require_grad_params_clip))
|
|
else:
|
|
params = itertools.chain(*self.lora_require_grad_params_clip)
|
|
|
|
|
|
# append paras
|
|
# ------------------------------------------------------------------
|
|
def add_cond_model(cond_model, params):
|
|
if isinstance(params[0], dict):
|
|
# parameter groups
|
|
params.append({"params": list(cond_model.parameters())})
|
|
else:
|
|
# parameter list: [torch.nn.parameter.Parameter]
|
|
params = params + list(cond_model.parameters())
|
|
return params
|
|
# ------------------------------------------------------------------
|
|
|
|
if self.cond_stage_trainable:
|
|
# print(f"{self.__class__.__name__}: Also optimizing conditioner params!")
|
|
params = add_cond_model(self.cond_stage_model, params)
|
|
|
|
if self.learn_logvar:
|
|
print('Diffusion model optimizing logvar')
|
|
if isinstance(params[0], dict):
|
|
params.append({"params": [self.logvar]})
|
|
else:
|
|
params.append(self.logvar)
|
|
|
|
# --------------------------------------------------------------------------------
|
|
opt = torch.optim.AdamW(params, lr=lr)
|
|
|
|
# lr scheduler
|
|
if self.use_scheduler:
|
|
assert 'target' in self.scheduler_config
|
|
scheduler = instantiate_from_config(self.scheduler_config)
|
|
|
|
print("Setting up LambdaLR scheduler...")
|
|
scheduler = [
|
|
{
|
|
'scheduler': LambdaLR(opt, lr_lambda=scheduler.schedule),
|
|
'interval': 'step',
|
|
'frequency': 1
|
|
}]
|
|
return [opt], scheduler
|
|
|
|
return opt
|
|
|
|
@torch.no_grad()
|
|
def to_rgb(self, x):
|
|
x = x.float()
|
|
if not hasattr(self, "colorize"):
|
|
self.colorize = torch.randn(3, x.shape[1], 1, 1).to(x)
|
|
x = nn.functional.conv2d(x, weight=self.colorize)
|
|
x = 2. * (x - x.min()) / (x.max() - x.min()) - 1.
|
|
return x
|
|
|
|
@torch.no_grad()
|
|
def mask_to_rgb(self, x):
|
|
x = x * 255
|
|
x = x.int()
|
|
return x
|
|
|
|
class DiffusionWrapper(pl.LightningModule):
|
|
def __init__(self, diff_model_config, conditioning_key):
|
|
super().__init__()
|
|
self.diffusion_model = instantiate_from_config(diff_model_config)
|
|
print('Successfully initialize the diffusion model !')
|
|
self.conditioning_key = conditioning_key
|
|
# assert self.conditioning_key in [None, 'concat', 'crossattn', 'hybrid', 'adm', 'resblockcond', 'hybrid-adm', 'hybrid-time']
|
|
|
|
def forward(self, x, t, c_concat: list = None, c_crossattn: list = None,
|
|
c_adm=None, s=None, mask=None, **kwargs):
|
|
# temporal_context = fps is foNone
|
|
if self.conditioning_key is None:
|
|
out = self.diffusion_model(x, t, **kwargs)
|
|
elif self.conditioning_key == 'concat':
|
|
xc = torch.cat([x] + c_concat, dim=1)
|
|
out = self.diffusion_model(xc, t, **kwargs)
|
|
elif self.conditioning_key == 'crossattn':
|
|
cc = torch.cat(c_crossattn, 1)
|
|
out = self.diffusion_model(x, t, context=cc, **kwargs)
|
|
elif self.conditioning_key == 'hybrid':
|
|
xc = torch.cat([x] + c_concat, dim=1)
|
|
cc = torch.cat(c_crossattn, 1)
|
|
out = self.diffusion_model(xc, t, context=cc, **kwargs)
|
|
elif self.conditioning_key == 'resblockcond':
|
|
cc = c_crossattn[0]
|
|
out = self.diffusion_model(x, t, context=cc, **kwargs)
|
|
elif self.conditioning_key == 'adm':
|
|
cc = c_crossattn[0]
|
|
out = self.diffusion_model(x, t, y=cc, **kwargs)
|
|
elif self.conditioning_key == 'hybrid-adm':
|
|
assert c_adm is not None
|
|
xc = torch.cat([x] + c_concat, dim=1)
|
|
cc = torch.cat(c_crossattn, 1)
|
|
out = self.diffusion_model(xc, t, context=cc, y=c_adm, **kwargs)
|
|
elif self.conditioning_key == 'hybrid-time':
|
|
assert s is not None
|
|
xc = torch.cat([x] + c_concat, dim=1)
|
|
cc = torch.cat(c_crossattn, 1)
|
|
out = self.diffusion_model(xc, t, context=cc, s=s, **kwargs)
|
|
elif self.conditioning_key == 'concat-time-mask':
|
|
# assert s is not None
|
|
# print('x & mask:',x.shape,c_concat[0].shape)
|
|
xc = torch.cat([x] + c_concat, dim=1)
|
|
out = self.diffusion_model(xc, t, context=None, s=s, mask=mask, **kwargs)
|
|
elif self.conditioning_key == 'concat-adm-mask':
|
|
# assert s is not None
|
|
# print('x & mask:',x.shape,c_concat[0].shape)
|
|
if c_concat is not None:
|
|
xc = torch.cat([x] + c_concat, dim=1)
|
|
else:
|
|
xc = x
|
|
out = self.diffusion_model(xc, t, context=None, y=s, mask=mask, **kwargs)
|
|
elif self.conditioning_key == 'crossattn-adm':
|
|
cc = torch.cat(c_crossattn, 1)
|
|
out = self.diffusion_model(x, t, context=cc, y=s, **kwargs)
|
|
elif self.conditioning_key == 'hybrid-adm-mask':
|
|
cc = torch.cat(c_crossattn, 1)
|
|
if c_concat is not None:
|
|
xc = torch.cat([x] + c_concat, dim=1)
|
|
else:
|
|
xc = x
|
|
out = self.diffusion_model(xc, t, context=cc, y=s, mask=mask, **kwargs)
|
|
elif self.conditioning_key == 'hybrid-time-adm': # adm means y, e.g., class index
|
|
# assert s is not None
|
|
assert c_adm is not None
|
|
xc = torch.cat([x] + c_concat, dim=1)
|
|
cc = torch.cat(c_crossattn, 1)
|
|
out = self.diffusion_model(xc, t, context=cc, s=s, y=c_adm, **kwargs)
|
|
else:
|
|
raise NotImplementedError()
|
|
|
|
return out
|
|
|
|
|
|
class T2VAdapterDepth(LatentDiffusion):
|
|
def __init__(self, depth_stage_config, adapter_config, *args, **kwargs):
|
|
super(T2VAdapterDepth, self).__init__(*args, **kwargs)
|
|
self.adapter = instantiate_from_config(adapter_config)
|
|
self.condtype = adapter_config.cond_name
|
|
self.depth_stage_model = instantiate_from_config(depth_stage_config)
|
|
|
|
def prepare_midas_input(self, batch_x):
|
|
# input: b,c,h,w
|
|
x_midas = torch.nn.functional.interpolate(batch_x, size=(384, 384), mode='bicubic')
|
|
return x_midas
|
|
|
|
@torch.no_grad()
|
|
def get_batch_depth(self, batch_x, target_size, encode_bs=1):
|
|
b, c, t, h, w = batch_x.shape
|
|
merge_x = rearrange(batch_x, 'b c t h w -> (b t) c h w')
|
|
split_x = torch.split(merge_x, encode_bs, dim=0)
|
|
cond_depth_list = []
|
|
for x in split_x:
|
|
x_midas = self.prepare_midas_input(x)
|
|
cond_depth = self.depth_stage_model(x_midas)
|
|
cond_depth = torch.nn.functional.interpolate(
|
|
cond_depth,
|
|
size=target_size,
|
|
mode="bicubic",
|
|
align_corners=False,
|
|
)
|
|
depth_min, depth_max = torch.amin(cond_depth, dim=[1, 2, 3], keepdim=True), torch.amax(cond_depth, dim=[1, 2, 3], keepdim=True)
|
|
cond_depth = 2. * (cond_depth - depth_min) / (depth_max - depth_min + 1e-7) - 1.
|
|
cond_depth_list.append(cond_depth)
|
|
batch_cond_depth=torch.cat(cond_depth_list, dim=0)
|
|
batch_cond_depth = rearrange(batch_cond_depth, '(b t) c h w -> b c t h w', b=b, t=t)
|
|
return batch_cond_depth
|
|
|
|
def get_adapter_features(self, extra_cond, encode_bs=1):
|
|
b, c, t, h, w = extra_cond.shape
|
|
## process in 2D manner
|
|
merge_extra_cond = rearrange(extra_cond, 'b c t h w -> (b t) c h w')
|
|
split_extra_cond = torch.split(merge_extra_cond, encode_bs, dim=0)
|
|
features_adapter_list = []
|
|
for extra_cond in split_extra_cond:
|
|
features_adapter = self.adapter(extra_cond)
|
|
features_adapter_list.append(features_adapter)
|
|
merge_features_adapter_list = []
|
|
for i in range(len(features_adapter_list[0])):
|
|
merge_features_adapter = torch.cat([features_adapter_list[num][i] for num in range(len(features_adapter_list))], dim=0)
|
|
merge_features_adapter_list.append(merge_features_adapter)
|
|
merge_features_adapter_list = [rearrange(feature, '(b t) c h w -> b c t h w', b=b, t=t) for feature in merge_features_adapter_list]
|
|
return merge_features_adapter_list
|