1685 lines
66 KiB
Python
1685 lines
66 KiB
Python
# Inspired by: https://github.com/Mikubill/sd-webui-controlnet/discussions/1236 and https://github.com/Mikubill/sd-webui-controlnet/discussions/1280
|
|
import inspect
|
|
import logging
|
|
from dataclasses import dataclass
|
|
from typing import Any, Callable, Dict, List, Optional, Tuple, Union
|
|
|
|
import numpy as np
|
|
import PIL.Image
|
|
import torch
|
|
from diffusers import StableDiffusionControlNetPipeline
|
|
from diffusers.image_processor import PipelineImageInput, VaeImageProcessor
|
|
from diffusers.models import ControlNetModel, UNet2DConditionModel
|
|
from diffusers.models.attention import BasicTransformerBlock
|
|
from diffusers.models.autoencoders import AutoencoderKL
|
|
from diffusers.models.unets.unet_2d_blocks import (
|
|
CrossAttnDownBlock2D,
|
|
CrossAttnUpBlock2D,
|
|
DownBlock2D,
|
|
UNetMidBlock2DCrossAttn,
|
|
UpBlock2D,
|
|
)
|
|
from diffusers.pipelines.controlnet.multicontrolnet import MultiControlNetModel
|
|
from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
|
|
from diffusers.pipelines.stable_diffusion.safety_checker import (
|
|
StableDiffusionSafetyChecker,
|
|
)
|
|
from diffusers.schedulers import KarrasDiffusionSchedulers
|
|
from diffusers.utils.torch_utils import is_compiled_module, randn_tensor
|
|
from transformers import (
|
|
CLIPImageProcessor,
|
|
CLIPTextModel,
|
|
CLIPTokenizer,
|
|
CLIPVisionModelWithProjection,
|
|
)
|
|
|
|
formatter = logging.Formatter("%(asctime)s - %(levelname)s - %(message)s")
|
|
logger = logging.getLogger(__name__)
|
|
logger.setLevel(logging.DEBUG)
|
|
ch = logging.StreamHandler()
|
|
ch.setFormatter(formatter)
|
|
logger.addHandler(ch)
|
|
|
|
basic_transformer_idx = 0
|
|
|
|
|
|
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
|
|
def retrieve_latents(
|
|
encoder_output: torch.Tensor,
|
|
generator: Optional[torch.Generator] = None,
|
|
sample_mode: str = "sample",
|
|
):
|
|
if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
|
|
return encoder_output.latent_dist.sample(generator)
|
|
elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
|
|
return encoder_output.latent_dist.mode()
|
|
elif hasattr(encoder_output, "latents"):
|
|
return encoder_output.latents
|
|
else:
|
|
raise AttributeError("Could not access latents of provided encoder_output")
|
|
|
|
|
|
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
|
|
def retrieve_timesteps(
|
|
scheduler,
|
|
num_inference_steps: Optional[int] = None,
|
|
device: Optional[Union[str, torch.device]] = None,
|
|
timesteps: Optional[List[int]] = None,
|
|
**kwargs,
|
|
):
|
|
"""
|
|
Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
|
|
custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
|
|
|
|
Args:
|
|
scheduler (`SchedulerMixin`):
|
|
The scheduler to get timesteps from.
|
|
num_inference_steps (`int`):
|
|
The number of diffusion steps used when generating samples with a pre-trained model. If used,
|
|
`timesteps` must be `None`.
|
|
device (`str` or `torch.device`, *optional*):
|
|
The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
|
|
timesteps (`List[int]`, *optional*):
|
|
Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default
|
|
timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
|
|
must be `None`.
|
|
|
|
Returns:
|
|
`Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
|
|
second element is the number of inference steps.
|
|
"""
|
|
if timesteps is not None:
|
|
accepts_timesteps = "timesteps" in set(
|
|
inspect.signature(scheduler.set_timesteps).parameters.keys()
|
|
)
|
|
if not accepts_timesteps:
|
|
raise ValueError(
|
|
f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
|
|
f" timestep schedules. Please check whether you are using the correct scheduler."
|
|
)
|
|
scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
|
|
timesteps = scheduler.timesteps
|
|
num_inference_steps = len(timesteps)
|
|
else:
|
|
scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
|
|
timesteps = scheduler.timesteps
|
|
return timesteps, num_inference_steps
|
|
|
|
|
|
def _images_to_tensors(
|
|
imgs: List[PIL.Image.Image],
|
|
width: int,
|
|
height: int,
|
|
device: torch.device,
|
|
dtype: torch.dtype,
|
|
) -> torch.Tensor:
|
|
buf = []
|
|
for image_ in imgs:
|
|
assert isinstance(image_, PIL.Image.Image)
|
|
image_ = image_.convert("RGB")
|
|
image_ = image_.resize((width, height), resample=PIL.Image.Resampling.LANCZOS)
|
|
image_ = np.array(image_)
|
|
image_ = image_[None, :]
|
|
buf.append(image_)
|
|
|
|
image = np.concatenate(buf, axis=0)
|
|
image = np.array(image).astype(np.float32) / 255.0
|
|
image = (image - 0.5) / 0.5
|
|
image = image.transpose(0, 3, 1, 2)
|
|
image = torch.from_numpy(image)
|
|
|
|
assert isinstance(image, torch.Tensor)
|
|
|
|
image = image.to(device=device, dtype=dtype)
|
|
|
|
return image
|
|
|
|
|
|
def mask_images_to_float_tensor(
|
|
imgs: List[PIL.Image.Image],
|
|
resize_wh: Optional[Tuple[int, int]] = None,
|
|
resample: Optional[PIL.Image.Resampling] = None,
|
|
) -> torch.Tensor:
|
|
width, height = imgs[0].size
|
|
if resize_wh is not None:
|
|
width, height = resize_wh
|
|
if resample is None:
|
|
resample = PIL.Image.Resampling.LANCZOS
|
|
mask = [i.resize((width, height), resample=resample) for i in imgs]
|
|
else:
|
|
mask = imgs
|
|
mask = np.stack([np.array(m.convert("L")) for m in mask], axis=0)
|
|
assert mask.shape == (len(imgs), height, width)
|
|
mask = mask.astype(np.float32) / 255.0
|
|
mask = torch.from_numpy(mask)
|
|
|
|
assert mask.shape[0] == len(imgs)
|
|
if mask.min() < 0 or mask.max() > 1:
|
|
raise ValueError("Mask should be in [0, 1] range")
|
|
return mask
|
|
|
|
|
|
def torch_dfs(model: torch.nn.Module):
|
|
result = [model]
|
|
for child in model.children():
|
|
result += torch_dfs(child)
|
|
return result
|
|
|
|
|
|
@dataclass
|
|
class ControlNetUnit:
|
|
controlnet: ControlNetModel
|
|
image: PIL.Image.Image
|
|
scale: float
|
|
start: float
|
|
end: float
|
|
|
|
|
|
class ControlNetUnits:
|
|
def __init__(
|
|
self,
|
|
units: tuple[ControlNetUnit],
|
|
):
|
|
self.controlnets = [unit.controlnet for unit in units]
|
|
self.images = [unit.image for unit in units]
|
|
self.scales = [unit.scale for unit in units]
|
|
self.starts = [unit.start for unit in units]
|
|
self.ends = [unit.end for unit in units]
|
|
|
|
|
|
class JannchiePipeline(StableDiffusionControlNetPipeline):
|
|
@torch.no_grad()
|
|
def __call__(
|
|
self,
|
|
prompt: Union[str, List[str]] = None,
|
|
image: Union[torch.FloatTensor, PIL.Image.Image] = None,
|
|
ref_image: Union[torch.FloatTensor, PIL.Image.Image] = None,
|
|
ref_image_mask: Union[torch.FloatTensor, PIL.Image.Image] = None,
|
|
height: Optional[int] = None,
|
|
width: Optional[int] = None,
|
|
num_inference_steps: int = 50,
|
|
guidance_scale: float = 7.5,
|
|
negative_prompt: Optional[Union[str, List[str]]] = None,
|
|
num_images_per_prompt: Optional[int] = 1,
|
|
eta: float = 0.0,
|
|
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
|
|
latents: Optional[torch.FloatTensor] = None,
|
|
prompt_embeds: Optional[torch.FloatTensor] = None,
|
|
negative_prompt_embeds: Optional[torch.FloatTensor] = None,
|
|
output_type: Optional[str] = "pil",
|
|
return_dict: bool = True,
|
|
callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
|
|
callback_steps: int = 1,
|
|
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
|
|
controlnet_units: ControlNetUnits = None,
|
|
guess_mode: bool = False,
|
|
reference_attn: bool = False,
|
|
reference_adain: bool = False,
|
|
attention_auto_machine_weight: float = 100.0,
|
|
gn_auto_machine_weight: float = 1.0,
|
|
style_fidelity: float = 0.5,
|
|
write_mask: Union[torch.FloatTensor, PIL.Image.Image] = None,
|
|
bool_mask=False,
|
|
desc: Optional[str] = None,
|
|
strength=1.0,
|
|
timesteps: List[int] = None,
|
|
mask_image: PipelineImageInput = None,
|
|
masked_image_latents: Optional[torch.FloatTensor] = None,
|
|
*arg,
|
|
**args,
|
|
):
|
|
device = self._execution_device
|
|
if height == None:
|
|
if isinstance(image, torch.Tensor):
|
|
if image is not None:
|
|
height = image.shape[-2]
|
|
elif ref_image is not None:
|
|
height = ref_image.shape[-2]
|
|
else:
|
|
height = 512
|
|
elif isinstance(image, PIL.Image.Image):
|
|
_, height = image.size
|
|
if width == None:
|
|
if isinstance(image, torch.Tensor):
|
|
if image is not None:
|
|
width = image.shape[-1]
|
|
elif ref_image is not None:
|
|
width = ref_image.shape[-1]
|
|
else:
|
|
width = 512
|
|
elif isinstance(image, PIL.Image.Image):
|
|
width, _ = image.size
|
|
|
|
if arg or args:
|
|
logger.warning(f"Unused arguments: {arg}, {args}")
|
|
if desc is None:
|
|
desc = "Jannchie's Pipeline"
|
|
self.set_progress_bar_config(desc=desc)
|
|
controlnet_conditioning_scale = []
|
|
control_guidance_start = []
|
|
control_guidance_end = []
|
|
if controlnet_units:
|
|
self.controlnet = MultiControlNetModel(
|
|
controlnets=controlnet_units.controlnets
|
|
)
|
|
controlnet_images = controlnet_units.images
|
|
control_guidance_start = controlnet_units.starts
|
|
control_guidance_end = controlnet_units.ends
|
|
controlnet_conditioning_scale = controlnet_units.scales
|
|
else:
|
|
controlnet_images = []
|
|
self.controlnet = MultiControlNetModel(controlnets=[])
|
|
if not reference_attn and not reference_adain:
|
|
ref_image = None
|
|
if self.controlnet:
|
|
controlnet = (
|
|
self.controlnet._orig_mod
|
|
if is_compiled_module(self.controlnet)
|
|
else self.controlnet
|
|
)
|
|
controlnet.to(device)
|
|
n_controlnet_unit = (
|
|
len(controlnet.nets)
|
|
if isinstance(controlnet, MultiControlNetModel)
|
|
else 0
|
|
)
|
|
else:
|
|
n_controlnet_unit = 0
|
|
# 1. Check inputs. Raise error if not correct
|
|
self.check_inputs(
|
|
prompt,
|
|
controlnet_images,
|
|
callback_steps,
|
|
negative_prompt,
|
|
prompt_embeds,
|
|
negative_prompt_embeds,
|
|
controlnet_conditioning_scale=controlnet_conditioning_scale,
|
|
control_guidance_start=control_guidance_start,
|
|
control_guidance_end=control_guidance_end,
|
|
)
|
|
|
|
# 2. Define call parameters
|
|
if prompt is not None and isinstance(prompt, str):
|
|
batch_size = 1
|
|
elif prompt is not None and isinstance(prompt, list):
|
|
batch_size = len(prompt)
|
|
else:
|
|
# 输入的是 prompt_embeds
|
|
batch_size = prompt_embeds.shape[0]
|
|
|
|
# here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
|
|
# of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
|
|
# corresponds to doing no classifier free guidance.
|
|
do_classifier_free_guidance = guidance_scale > 1.0
|
|
if self.controlnet:
|
|
if len(self.controlnet.nets) > 1:
|
|
assert isinstance(controlnet_images, list)
|
|
if n_controlnet_unit != 0:
|
|
global_pool_conditions = (
|
|
controlnet.config.global_pool_conditions
|
|
if isinstance(controlnet, ControlNetModel)
|
|
else controlnet.nets[0].config.global_pool_conditions
|
|
)
|
|
guess_mode = guess_mode or global_pool_conditions
|
|
|
|
# 3. Encode input prompt
|
|
logger.debug("Encoding prompt")
|
|
text_encoder_lora_scale = (
|
|
cross_attention_kwargs.get("scale", None)
|
|
if cross_attention_kwargs is not None
|
|
else None
|
|
)
|
|
prompt_embeds = self.encode_prompt(
|
|
prompt,
|
|
device,
|
|
num_images_per_prompt,
|
|
do_classifier_free_guidance,
|
|
negative_prompt,
|
|
prompt_embeds=prompt_embeds,
|
|
negative_prompt_embeds=negative_prompt_embeds,
|
|
lora_scale=text_encoder_lora_scale,
|
|
)
|
|
prompt_embeds = torch.cat(prompt_embeds[::-1], dim=0)
|
|
|
|
# 4. Prepare image
|
|
logger.debug("Preparing image")
|
|
if n_controlnet_unit != 0:
|
|
if isinstance(controlnet, ControlNetModel):
|
|
controlnet_images = self.prepare_image(
|
|
image=controlnet_images,
|
|
width=width,
|
|
height=height,
|
|
batch_size=batch_size * num_images_per_prompt,
|
|
num_images_per_prompt=num_images_per_prompt,
|
|
device=device,
|
|
dtype=controlnet.dtype,
|
|
do_classifier_free_guidance=do_classifier_free_guidance,
|
|
guess_mode=guess_mode,
|
|
).to(device=device)
|
|
height, width = controlnet_images.shape[-2:]
|
|
elif isinstance(controlnet, MultiControlNetModel):
|
|
images = []
|
|
|
|
for image_ in controlnet_images:
|
|
image_ = self.prepare_image(
|
|
image=image_,
|
|
width=width,
|
|
height=height,
|
|
batch_size=batch_size * num_images_per_prompt,
|
|
num_images_per_prompt=num_images_per_prompt,
|
|
device=device,
|
|
dtype=controlnet.dtype,
|
|
do_classifier_free_guidance=do_classifier_free_guidance,
|
|
guess_mode=guess_mode,
|
|
).to(device=device)
|
|
|
|
images.append(image_)
|
|
|
|
controlnet_images = images
|
|
height, width = controlnet_images[0].shape[-2:]
|
|
else:
|
|
assert False
|
|
|
|
# 5. Preprocess reference image
|
|
logger.debug("Preprocessing reference image")
|
|
if ref_image is not None:
|
|
if isinstance(ref_image, PIL.Image.Image):
|
|
ref_image = self.image_processor.preprocess(
|
|
ref_image, height=height, width=width
|
|
)
|
|
ref_image = self.norm_image_tensor(ref_image)
|
|
# 6. Prepare timesteps
|
|
logger.debug("Preparing timesteps")
|
|
timesteps, num_inference_steps = retrieve_timesteps(
|
|
self.scheduler, num_inference_steps, device, timesteps
|
|
)
|
|
timesteps, num_inference_steps = self.get_timesteps(
|
|
num_inference_steps=num_inference_steps,
|
|
strength=strength,
|
|
)
|
|
latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
|
|
|
|
# 7. Prepare latent variables
|
|
logger.debug("Preparing latent variables")
|
|
num_channels_latents = self.unet.config.in_channels
|
|
if image is not None:
|
|
if isinstance(image, PIL.Image.Image):
|
|
image = self.image_processor.preprocess(
|
|
image, height=height, width=width
|
|
)
|
|
if isinstance(image, torch.Tensor):
|
|
image = self.norm_image_tensor(image)
|
|
input_latents = self.image_to_latents(
|
|
image,
|
|
batch_size * num_images_per_prompt,
|
|
self.unet.dtype,
|
|
device,
|
|
generator,
|
|
False, # it will duplicate the latents after this step
|
|
)
|
|
|
|
logger.debug("Preparing latents")
|
|
num_channels_unet = self.unet.config.in_channels
|
|
return_image_latents = num_channels_unet == 4
|
|
latents_outputs = self.prepare_latents(
|
|
batch_size * num_images_per_prompt,
|
|
num_channels_latents,
|
|
height,
|
|
width,
|
|
self.unet.dtype,
|
|
device,
|
|
generator,
|
|
latents,
|
|
image,
|
|
latent_timestep,
|
|
is_strength_max=strength == 1.0,
|
|
return_noise=True,
|
|
return_image_latents=return_image_latents,
|
|
)
|
|
if return_image_latents:
|
|
input_latents, noise, image_latents = latents_outputs
|
|
else:
|
|
input_latents, noise = latents_outputs
|
|
|
|
# 7. Prepare mask latent variables
|
|
if mask_image is not None:
|
|
mask_condition = self.mask_processor.preprocess(
|
|
mask_image, height=height, width=width
|
|
)
|
|
init_image = image
|
|
init_image = init_image.to(dtype=torch.float32)
|
|
if masked_image_latents is None:
|
|
masked_image = init_image * (mask_condition < 0.5)
|
|
else:
|
|
masked_image = masked_image_latents
|
|
mask, masked_image_latents = self.prepare_mask_latents(
|
|
mask_condition,
|
|
masked_image,
|
|
batch_size * num_images_per_prompt,
|
|
height,
|
|
width,
|
|
prompt_embeds.dtype,
|
|
device,
|
|
generator,
|
|
do_classifier_free_guidance,
|
|
)
|
|
|
|
# 8. Prepare reference latent variables
|
|
if ref_image is not None:
|
|
ref_image_latents = self.image_to_latents(
|
|
ref_image,
|
|
batch_size * num_images_per_prompt,
|
|
self.unet.dtype,
|
|
device,
|
|
generator,
|
|
do_classifier_free_guidance,
|
|
)
|
|
|
|
# 9. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
|
|
extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
|
|
|
|
ref_mask_dict, out_mask_dict = self.get_ref_mask_dicts(
|
|
ref_image_mask,
|
|
height,
|
|
width,
|
|
num_images_per_prompt,
|
|
write_mask,
|
|
bool_mask,
|
|
device,
|
|
batch_size,
|
|
)
|
|
ref_data = ReferenceData(
|
|
ref_image=ref_image,
|
|
ref_image_mask=ref_image_mask,
|
|
style_fidelity=style_fidelity,
|
|
attention_auto_machine_weight=attention_auto_machine_weight,
|
|
gn_auto_machine_weight=gn_auto_machine_weight,
|
|
ref_mask_dict=ref_mask_dict,
|
|
out_mask_dict=out_mask_dict,
|
|
)
|
|
if reference_attn:
|
|
self.unet = ReferenceOnlyUNet2DConditionModel.from_unet(
|
|
self.unet,
|
|
ref_data,
|
|
reference_attn,
|
|
reference_adain,
|
|
)
|
|
else:
|
|
self.unet = ReferenceOnlyUNet2DConditionModel.revert_unet(
|
|
self.unet
|
|
) # 9. Modify self attention and group norm
|
|
if ref_image is not None:
|
|
self.unet.ref_data.MODE = "write"
|
|
self.unet.ref_data.uc_mask = (
|
|
torch.Tensor(
|
|
[1] * batch_size * num_images_per_prompt
|
|
+ [0] * batch_size * num_images_per_prompt
|
|
)
|
|
.type_as(ref_image_latents)
|
|
.bool()
|
|
)
|
|
|
|
if self.controlnet:
|
|
# Create tensor stating which controlnets to keep
|
|
controlnet_keep = []
|
|
for i in range(len(timesteps)):
|
|
keeps = [
|
|
1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
|
|
for s, e in zip(control_guidance_start, control_guidance_end)
|
|
]
|
|
controlnet_keep.append(
|
|
keeps[0] if isinstance(controlnet, ControlNetModel) else keeps
|
|
)
|
|
|
|
# 11. Denoising loop
|
|
num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
|
|
with self.progress_bar(total=num_inference_steps) as progress_bar:
|
|
for i, t in enumerate(timesteps):
|
|
# ref only part
|
|
if reference_attn:
|
|
self.unet.ref_data.progress = i / num_inference_steps
|
|
|
|
if ref_image is not None:
|
|
single_shape = (1,) + ref_image_latents.shape[1:]
|
|
single_noise = randn_tensor(
|
|
single_shape,
|
|
generator=generator,
|
|
device=device,
|
|
dtype=ref_image_latents.dtype,
|
|
)
|
|
noise_for_ref = single_noise.repeat_interleave(
|
|
ref_image_latents.shape[0], dim=0
|
|
)
|
|
ref_xt = self.scheduler.add_noise(
|
|
ref_image_latents,
|
|
noise_for_ref,
|
|
t.reshape(
|
|
1,
|
|
),
|
|
)
|
|
# ref_xt = self.scheduler.scale_model_input(ref_xt, t)
|
|
|
|
self.unet.ref_data.MODE = "write"
|
|
self.unet(
|
|
ref_xt,
|
|
t,
|
|
encoder_hidden_states=prompt_embeds,
|
|
cross_attention_kwargs=cross_attention_kwargs,
|
|
return_dict=False,
|
|
)
|
|
self.unet.ref_data.MODE = "read"
|
|
|
|
# expand the latents if we are doing classifier free guidance
|
|
latent_model_input = (
|
|
torch.cat([input_latents] * 2)
|
|
if do_classifier_free_guidance
|
|
else input_latents
|
|
)
|
|
|
|
# controlnet(s) inference
|
|
if guess_mode and do_classifier_free_guidance:
|
|
# Infer ControlNet only for the conditional batch.
|
|
control_model_input = input_latents
|
|
control_model_input = self.scheduler.scale_model_input(
|
|
control_model_input, t
|
|
)
|
|
controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
|
|
else:
|
|
control_model_input = latent_model_input
|
|
controlnet_prompt_embeds = prompt_embeds
|
|
|
|
# calculate final conditioning_scale
|
|
if isinstance(controlnet_keep[i], list):
|
|
cond_scale = [
|
|
c * s
|
|
for c, s in zip(
|
|
controlnet_conditioning_scale, controlnet_keep[i]
|
|
)
|
|
]
|
|
else:
|
|
controlnet_cond_scale = controlnet_conditioning_scale
|
|
if isinstance(controlnet_cond_scale, list):
|
|
controlnet_cond_scale = controlnet_cond_scale[0]
|
|
cond_scale = controlnet_cond_scale * controlnet_keep[i]
|
|
|
|
assert isinstance(
|
|
self.controlnet, (ControlNetModel, MultiControlNetModel)
|
|
)
|
|
if n_controlnet_unit != 0:
|
|
down_block_res_samples, mid_block_res_sample = self.controlnet(
|
|
control_model_input,
|
|
t,
|
|
encoder_hidden_states=controlnet_prompt_embeds,
|
|
controlnet_cond=controlnet_images,
|
|
conditioning_scale=cond_scale,
|
|
guess_mode=guess_mode,
|
|
return_dict=False,
|
|
)
|
|
|
|
if guess_mode and do_classifier_free_guidance:
|
|
# Infered ControlNet only for the conditional batch.
|
|
# To apply the output of ControlNet to both the unconditional and conditional batches,
|
|
# add 0 to the unconditional batch to keep it unchanged.
|
|
down_block_res_samples = [
|
|
torch.cat([torch.zeros_like(d), d])
|
|
for d in down_block_res_samples
|
|
]
|
|
mid_block_res_sample = torch.cat(
|
|
[
|
|
torch.zeros_like(mid_block_res_sample),
|
|
mid_block_res_sample,
|
|
]
|
|
)
|
|
else:
|
|
down_block_res_samples, mid_block_res_sample = None, None
|
|
# predict the noise residual
|
|
noise_pred = self.unet(
|
|
latent_model_input,
|
|
t,
|
|
encoder_hidden_states=prompt_embeds,
|
|
cross_attention_kwargs=cross_attention_kwargs,
|
|
down_block_additional_residuals=down_block_res_samples,
|
|
mid_block_additional_residual=mid_block_res_sample,
|
|
)["sample"]
|
|
# perform guidance
|
|
if do_classifier_free_guidance:
|
|
noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
|
|
noise_pred = noise_pred_uncond + guidance_scale * (
|
|
noise_pred_text - noise_pred_uncond
|
|
)
|
|
|
|
# compute the previous noisy sample x_t -> x_t-1
|
|
input_latents = self.scheduler.step(
|
|
noise_pred, t, input_latents, **extra_step_kwargs
|
|
)["prev_sample"]
|
|
if num_channels_unet == 4 and (
|
|
mask_image is not None or masked_image_latents is not None
|
|
):
|
|
init_latents_proper = image_latents
|
|
if do_classifier_free_guidance:
|
|
init_mask, _ = mask.chunk(2)
|
|
else:
|
|
init_mask = mask
|
|
|
|
if i < len(timesteps) - 1:
|
|
noise_timestep = timesteps[i + 1]
|
|
init_latents_proper = self.scheduler.add_noise(
|
|
init_latents_proper, noise, torch.tensor([noise_timestep])
|
|
)
|
|
|
|
input_latents = (
|
|
1 - init_mask
|
|
) * init_latents_proper + init_mask * input_latents
|
|
# call the callback, if provided
|
|
if i == len(timesteps) - 1 or (
|
|
(i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0
|
|
):
|
|
progress_bar.update()
|
|
if callback is not None and i % callback_steps == 0:
|
|
step_idx = i // getattr(self.scheduler, "order", 1)
|
|
callback(step_idx, t, input_latents)
|
|
# If we do sequential model offloading, let's offload unet and controlnet
|
|
# manually for max memory savings
|
|
if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
|
|
self.unet.to("cpu")
|
|
self.controlnet.to("cpu")
|
|
torch.cuda.empty_cache()
|
|
|
|
if output_type != "latent":
|
|
result_imgs = self.vae.decode(
|
|
input_latents / self.vae.config.scaling_factor, return_dict=False
|
|
)[0]
|
|
result_imgs, has_nsfw_concept = self.run_safety_checker(
|
|
result_imgs, device, prompt_embeds.dtype
|
|
)
|
|
else:
|
|
result_imgs = input_latents
|
|
has_nsfw_concept = None
|
|
|
|
if has_nsfw_concept is None:
|
|
do_denormalize = [True] * result_imgs.shape[0]
|
|
else:
|
|
if isinstance(has_nsfw_concept, list):
|
|
do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
|
|
else:
|
|
do_denormalize = [not has_nsfw_concept]
|
|
# nan to zero
|
|
result_imgs = torch.nan_to_num(result_imgs, nan=0.0, posinf=0.0, neginf=0.0)
|
|
result_imgs = self.image_processor.postprocess(
|
|
result_imgs, output_type=output_type, do_denormalize=do_denormalize
|
|
)
|
|
|
|
# Offload last model to CPU
|
|
if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
|
|
self.final_offload_hook.offload()
|
|
|
|
if not return_dict:
|
|
return (result_imgs, has_nsfw_concept)
|
|
img_out = self.get_img_from_latents(latents=input_latents)
|
|
|
|
modules = torch_dfs(self.unet)
|
|
# 卸载 ref only hack
|
|
for module in modules:
|
|
|
|
if getattr(module, "_original_inner_forward", None) is not None:
|
|
# unregister the attn forward hook
|
|
module.forward = module._original_inner_forward
|
|
|
|
if getattr(module, "original_forward", None) is not None:
|
|
# unregister the adain forward hook
|
|
module.forward = module.original_forward
|
|
|
|
return StableDiffusionPipelineOutput(
|
|
images=img_out, nsfw_content_detected=has_nsfw_concept
|
|
)
|
|
|
|
def get_ref_mask_dicts(
|
|
self,
|
|
ref_image_mask,
|
|
height,
|
|
width,
|
|
num_images_per_prompt,
|
|
write_mask,
|
|
bool_mask,
|
|
device,
|
|
batch_size,
|
|
):
|
|
latent_width = width // self.vae_scale_factor
|
|
latent_height = height // self.vae_scale_factor
|
|
ref_mask_dict = {}
|
|
out_mask_dict = {}
|
|
for i in range(4):
|
|
w = latent_width >> i
|
|
h = latent_height >> i
|
|
|
|
resize_wh = (w, h)
|
|
if ref_image_mask:
|
|
# resize ref_iamge_mask
|
|
tmp_mt_key = mask_images_to_float_tensor(
|
|
[ref_image_mask],
|
|
resize_wh=resize_wh,
|
|
).to(device=device, dtype=self.unet.dtype)
|
|
|
|
mt_key = (
|
|
tmp_mt_key.flatten() > 0.5 if bool_mask else tmp_mt_key.flatten()
|
|
)
|
|
ref_mask_dict[mt_key.shape[-1]] = mt_key.repeat(
|
|
batch_size * num_images_per_prompt, 1
|
|
)
|
|
|
|
if write_mask:
|
|
tmp_mt_query = mask_images_to_float_tensor(
|
|
[write_mask],
|
|
resize_wh=resize_wh,
|
|
).to(device=device, dtype=self.unet.dtype)
|
|
if bool_mask:
|
|
mt_query = tmp_mt_query.flatten(1) > 0.5
|
|
else:
|
|
mt_query = tmp_mt_query.flatten(1)
|
|
out_mask_dict[mt_query.shape[-1]] = mt_query.repeat(
|
|
batch_size * num_images_per_prompt, 1
|
|
)
|
|
|
|
return ref_mask_dict, out_mask_dict
|
|
|
|
def norm_image_tensor(self, ref_image):
|
|
# 如果 image 维度为 3，说明没有 batch 维度
|
|
if len(ref_image.shape) == 3:
|
|
# 增加 batch 维度
|
|
ref_image = ref_image.unsqueeze(0)
|
|
if ref_image.shape[3] == 3:
|
|
# 转换成 channel 在前的形式
|
|
ref_image = ref_image.permute(0, 3, 1, 2)
|
|
if ref_image.min() >= 0:
|
|
# 数值为 0 ~ 1
|
|
# 数值规范到 -1 ~ 1
|
|
ref_image = (ref_image * 2 - 1).clamp(-1, 1)
|
|
return ref_image
|
|
|
|
def __init__(
|
|
self,
|
|
vae: AutoencoderKL,
|
|
text_encoder: CLIPTextModel,
|
|
tokenizer: CLIPTokenizer,
|
|
unet: UNet2DConditionModel,
|
|
scheduler: KarrasDiffusionSchedulers,
|
|
safety_checker: StableDiffusionSafetyChecker,
|
|
feature_extractor: CLIPImageProcessor,
|
|
controlnet: Union[
|
|
ControlNetModel,
|
|
List[ControlNetModel],
|
|
Tuple[ControlNetModel],
|
|
MultiControlNetModel,
|
|
] = None,
|
|
image_encoder: CLIPVisionModelWithProjection = None,
|
|
requires_safety_checker: bool = True,
|
|
):
|
|
if controlnet is None:
|
|
controlnet = []
|
|
pipe_class_name = self.__class__.__name__
|
|
self.set_progress_bar_config(
|
|
desc=f"Running {pipe_class_name}...",
|
|
unit_scale=True,
|
|
bar_format="{desc}: {percentage:3.0f}%|{bar}| {n_fmt}/{total_fmt} [{elapsed}<{remaining}, {rate_fmt}{postfix}]",
|
|
)
|
|
self.vae: AutoencoderKL
|
|
self.text_encoder: CLIPTextModel
|
|
self.tokenizer: CLIPTokenizer
|
|
self.unet: UNet2DConditionModel
|
|
super().__init__(
|
|
vae=vae,
|
|
text_encoder=text_encoder,
|
|
tokenizer=tokenizer,
|
|
unet=unet,
|
|
controlnet=controlnet,
|
|
scheduler=scheduler,
|
|
safety_checker=safety_checker,
|
|
feature_extractor=feature_extractor,
|
|
image_encoder=image_encoder,
|
|
requires_safety_checker=requires_safety_checker,
|
|
)
|
|
self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
|
|
self.mask_processor = VaeImageProcessor(
|
|
vae_scale_factor=self.vae_scale_factor,
|
|
do_normalize=False,
|
|
do_binarize=True,
|
|
do_convert_grayscale=True,
|
|
)
|
|
|
|
def image_to_latents(
|
|
self,
|
|
image,
|
|
batch_size,
|
|
dtype,
|
|
device,
|
|
generator,
|
|
do_classifier_free_guidance,
|
|
):
|
|
image = image.to(device=device, dtype=dtype)
|
|
|
|
# encode the mask image into latents space so we can concatenate it to the latents
|
|
if isinstance(generator, list):
|
|
image_latents = [
|
|
self.vae.encode(image[i : i + 1]).latent_dist.sample(
|
|
generator=generator[i]
|
|
)
|
|
for i in range(batch_size)
|
|
]
|
|
image_latents = torch.cat(image_latents, dim=0)
|
|
else:
|
|
image_latents = self.vae.encode(image).latent_dist.sample(
|
|
generator=generator
|
|
)
|
|
image_latents = self.vae.config.scaling_factor * image_latents
|
|
|
|
# duplicate mask and ref_image_latents for each generation per prompt, using mps friendly method
|
|
if image_latents.shape[0] < batch_size:
|
|
if batch_size % image_latents.shape[0] != 0:
|
|
raise ValueError(
|
|
"The passed images and the required batch size don't match. Images are supposed to be duplicated"
|
|
f" to a total batch size of {batch_size}, but {image_latents.shape[0]} images were passed."
|
|
" Make sure the number of images that you pass is divisible by the total requested batch size."
|
|
)
|
|
image_latents = image_latents.repeat(
|
|
batch_size // image_latents.shape[0], 1, 1, 1
|
|
)
|
|
|
|
image_latents = (
|
|
torch.cat([image_latents] * 2)
|
|
if do_classifier_free_guidance
|
|
else image_latents
|
|
)
|
|
|
|
# aligning device to prevent device errors when concating it with the latent model input
|
|
image_latents = image_latents.to(device=device, dtype=dtype)
|
|
return image_latents
|
|
|
|
# def decode_latents(self, latents: torch.Tensor):
|
|
# return self.get_img_from_latents(latents=latents)
|
|
|
|
def get_img_from_latents(self, latents: torch.Tensor):
|
|
# scale and decode the image latents with vae
|
|
if len(latents.shape) == 3:
|
|
latents = latents[None]
|
|
norm_latents = latents
|
|
dec_tensor = self.vae.decode(
|
|
norm_latents / self.vae.config.scaling_factor, return_dict=False
|
|
)[0]
|
|
dec_images = self.image_processor.postprocess(
|
|
dec_tensor, output_type="np", do_denormalize=[True] * dec_tensor.shape[0]
|
|
)
|
|
dec_image_zero = dec_images
|
|
dec_image_zero = np.nan_to_num(dec_image_zero)
|
|
image_out_np = np.clip(
|
|
(dec_image_zero * 255.0).round().astype(int), a_min=0, a_max=255
|
|
).astype(np.uint8)
|
|
return [PIL.Image.fromarray(img) for img in image_out_np]
|
|
|
|
def encode_images_to_latents(
|
|
self,
|
|
imgs: List[PIL.Image.Image],
|
|
generator: torch.Generator,
|
|
device: torch.device,
|
|
dtype: torch.dtype,
|
|
) -> torch.Tensor:
|
|
width, height = imgs[0].size
|
|
image_tensor = _images_to_tensors(
|
|
imgs=imgs, width=width, height=height, device=device, dtype=dtype
|
|
)
|
|
# encode the mask image into latents space so we can concatenate it to the latents
|
|
if isinstance(generator, list):
|
|
image_latent = torch.cat(
|
|
[
|
|
self.vae.encode(image_tensor[i : i + 1]).latent_dist.sample(
|
|
generator=generator[i]
|
|
)
|
|
for i in range(image_tensor.shape[0])
|
|
],
|
|
dim=0,
|
|
)
|
|
else:
|
|
image_latent = self.vae.encode(image_tensor).latent_dist.sample(
|
|
generator=generator
|
|
)
|
|
image_latent = self.vae.config.scaling_factor * image_latent
|
|
|
|
return image_latent.to(device=device, dtype=dtype)
|
|
|
|
def get_timesteps(self, num_inference_steps, strength: float):
|
|
# get the original timestep using init_timestep
|
|
init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
|
|
t_start = max(num_inference_steps - init_timestep, 0)
|
|
timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
|
|
if hasattr(self.scheduler, "set_begin_index"):
|
|
self.scheduler.set_begin_index(t_start * self.scheduler.order)
|
|
|
|
return timesteps, num_inference_steps - t_start
|
|
|
|
def prepare_latents(
|
|
self,
|
|
batch_size,
|
|
num_channels_latents,
|
|
height,
|
|
width,
|
|
dtype,
|
|
device,
|
|
generator,
|
|
latents=None,
|
|
image=None,
|
|
timestep=None,
|
|
is_strength_max=True,
|
|
return_noise=False,
|
|
return_image_latents=False,
|
|
):
|
|
shape = (
|
|
batch_size,
|
|
num_channels_latents,
|
|
height // self.vae_scale_factor,
|
|
width // self.vae_scale_factor,
|
|
)
|
|
if return_image_latents or (latents is None and not is_strength_max):
|
|
# TODO: check it
|
|
image = image.to(device=device, dtype=dtype)
|
|
|
|
if image.shape[1] == 4:
|
|
image_latents = image
|
|
else:
|
|
image_latents = self._encode_vae_image(image=image, generator=generator)
|
|
image_latents = image_latents.repeat(
|
|
batch_size // image_latents.shape[0], 1, 1, 1
|
|
)
|
|
|
|
if latents is None:
|
|
noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
|
|
if is_strength_max:
|
|
latents = noise * self.scheduler.init_noise_sigma
|
|
else:
|
|
latents = self.scheduler.add_noise(image_latents, noise, timestep)
|
|
else:
|
|
noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
|
|
if is_strength_max:
|
|
latents = noise * self.scheduler.init_noise_sigma
|
|
else:
|
|
latents = self.scheduler.add_noise(latents, noise, timestep)
|
|
|
|
outputs = (latents,)
|
|
|
|
if return_noise:
|
|
outputs += (noise,)
|
|
|
|
if return_image_latents:
|
|
outputs += (image_latents,)
|
|
return outputs
|
|
|
|
def prepare_mask_latents(
|
|
self,
|
|
mask: torch.Tensor,
|
|
masked_image: torch.Tensor,
|
|
batch_size: int,
|
|
height: int,
|
|
width: int,
|
|
dtype: torch.dtype,
|
|
device: torch.device,
|
|
generator: torch.Generator,
|
|
do_classifier_free_guidance: bool,
|
|
):
|
|
# resize the mask to latents shape as we concatenate the mask to the latents
|
|
# we do that before converting to dtype to avoid breaking in case we're using cpu_offload
|
|
# and half precision
|
|
mask = torch.nn.functional.interpolate(
|
|
mask, size=(height // self.vae_scale_factor, width // self.vae_scale_factor)
|
|
)
|
|
mask = mask.to(device=device, dtype=dtype)
|
|
|
|
masked_image = masked_image.to(device=device, dtype=dtype)
|
|
|
|
if masked_image.shape[1] == 4:
|
|
masked_image_latents = masked_image
|
|
else:
|
|
masked_image_latents = self._encode_vae_image(
|
|
masked_image, generator=generator
|
|
)
|
|
|
|
# duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
|
|
if mask.shape[0] < batch_size:
|
|
if batch_size % mask.shape[0] != 0:
|
|
raise ValueError(
|
|
"The passed mask and the required batch size don't match. Masks are supposed to be duplicated to"
|
|
f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number"
|
|
" of masks that you pass is divisible by the total requested batch size."
|
|
)
|
|
mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1)
|
|
if masked_image_latents.shape[0] < batch_size:
|
|
if batch_size % masked_image_latents.shape[0] != 0:
|
|
raise ValueError(
|
|
"The passed images and the required batch size don't match. Images are supposed to be duplicated"
|
|
f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed."
|
|
" Make sure the number of images that you pass is divisible by the total requested batch size."
|
|
)
|
|
masked_image_latents = masked_image_latents.repeat(
|
|
batch_size // masked_image_latents.shape[0], 1, 1, 1
|
|
)
|
|
|
|
mask = torch.cat([mask] * 2) if do_classifier_free_guidance else mask
|
|
masked_image_latents = (
|
|
torch.cat([masked_image_latents] * 2)
|
|
if do_classifier_free_guidance
|
|
else masked_image_latents
|
|
)
|
|
|
|
# aligning device to prevent device errors when concating it with the latent model input
|
|
masked_image_latents = masked_image_latents.to(device=device, dtype=dtype)
|
|
return mask, masked_image_latents
|
|
|
|
def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator):
|
|
if isinstance(generator, list):
|
|
image_latents = [
|
|
retrieve_latents(
|
|
self.vae.encode(image[i : i + 1]), generator=generator[i]
|
|
)
|
|
for i in range(image.shape[0])
|
|
]
|
|
image_latents = torch.cat(image_latents, dim=0)
|
|
else:
|
|
image_latents = retrieve_latents(
|
|
self.vae.encode(image), generator=generator
|
|
)
|
|
|
|
image_latents = self.vae.config.scaling_factor * image_latents
|
|
return image_latents
|
|
|
|
|
|
@dataclass
|
|
class ReferenceData:
|
|
ref_image: Union[torch.FloatTensor, PIL.Image.Image] = None
|
|
ref_image_mask: Union[torch.FloatTensor, PIL.Image.Image] = None
|
|
MODE: str = "write"
|
|
progress: float = 0.0
|
|
uc_mask: torch.Tensor = None
|
|
bool_mask: bool = True
|
|
style_fidelity: float = 1.0
|
|
do_classifier_free_guidance: bool = True
|
|
attention_auto_machine_weight: float = 100.0
|
|
gn_auto_machine_weight: float = 1.0
|
|
ref_mask_dict: dict = None
|
|
out_mask_dict: dict = None
|
|
|
|
|
|
class ReferenceOnlyUNet2DConditionModel(UNet2DConditionModel):
|
|
@classmethod
|
|
def from_unet(
|
|
cls,
|
|
unet: UNet2DConditionModel,
|
|
ref_data: ReferenceData = ReferenceData(),
|
|
reference_attn: bool = False,
|
|
reference_adain: bool = False,
|
|
) -> "ReferenceOnlyUNet2DConditionModel":
|
|
# 创建一个新的子类实例
|
|
basic_transformer_idx = 0
|
|
basic_transformer_blocks = []
|
|
for module in torch_dfs(unet):
|
|
if reference_attn:
|
|
if isinstance(module, BasicTransformerBlock):
|
|
basic_transformer_blocks.append(module)
|
|
module.__class__ = BasicTransformerBlockReferenceOnly
|
|
module.ref_data = ref_data
|
|
module.bank = []
|
|
module.idx = basic_transformer_idx
|
|
basic_transformer_idx += 1
|
|
elif reference_adain:
|
|
if isinstance(module, CrossAttnDownBlock2D):
|
|
module.__class__ = CrossAttnDownBlock2DReferenceOnly
|
|
module.ref_data = ref_data
|
|
module.bank = []
|
|
if isinstance(module, DownBlock2D):
|
|
module.__class__ = DownBlock2DReferenceOnly
|
|
if isinstance(module, UNetMidBlock2DCrossAttn):
|
|
module.__class__ = UNetMidBlock2DCrossAttnReferenceOnly
|
|
if isinstance(module, UpBlock2D):
|
|
module.__class__ = UpBlock2DReferenceOnly
|
|
if isinstance(module, CrossAttnUpBlock2D):
|
|
module.__class__ = CrossAttnUpBlock2DReferenceOnly
|
|
module.ref_data = ref_data
|
|
unet.mid_block.gn_weight = 0
|
|
down_blocks = unet.down_blocks
|
|
module.mean_bank = []
|
|
module.var_bank = []
|
|
for w, module in enumerate(down_blocks):
|
|
module.gn_weight = 1.0 - float(w) / float(len(down_blocks))
|
|
module.gn_weight *= 2
|
|
|
|
up_blocks = unet.up_blocks
|
|
for w, module in enumerate(up_blocks):
|
|
module.gn_weight = float(w) / float(len(up_blocks))
|
|
module.gn_weight *= 2
|
|
|
|
# 计算 attn_weight
|
|
basic_transformer_blocks = sorted(
|
|
basic_transformer_blocks, key=lambda x: -x.norm1.normalized_shape[0]
|
|
)
|
|
|
|
for i, module in enumerate(basic_transformer_blocks):
|
|
module.attn_weight = float(i) / float(len(basic_transformer_blocks))
|
|
unet.__class__ = cls
|
|
unet.ref_data = ref_data
|
|
return unet
|
|
|
|
@classmethod
|
|
def revert_unet(
|
|
cls, unet: "ReferenceOnlyUNet2DConditionModel"
|
|
) -> UNet2DConditionModel:
|
|
unet.__class__ = UNet2DConditionModel
|
|
for module in torch_dfs(unet):
|
|
if isinstance(module, BasicTransformerBlockReferenceOnly):
|
|
module.__class__ = BasicTransformerBlock
|
|
if isinstance(module, CrossAttnDownBlock2DReferenceOnly):
|
|
module.__class__ = CrossAttnDownBlock2D
|
|
if isinstance(module, DownBlock2DReferenceOnly):
|
|
module.__class__ = DownBlock2D
|
|
if isinstance(module, UNetMidBlock2DCrossAttnReferenceOnly):
|
|
module.__class__ = UNetMidBlock2DCrossAttn
|
|
if isinstance(module, UpBlock2DReferenceOnly):
|
|
module.__class__ = UpBlock2D
|
|
if isinstance(module, CrossAttnUpBlock2DReferenceOnly):
|
|
module.__class__ = CrossAttnUpBlock2D
|
|
return unet
|
|
|
|
|
|
class BasicTransformerBlockReferenceOnly(BasicTransformerBlock):
|
|
|
|
@classmethod
|
|
def from_module(
|
|
cls, module: BasicTransformerBlock
|
|
) -> "BasicTransformerBlockReferenceOnly":
|
|
module.__class__ = cls
|
|
return module
|
|
|
|
def forward(
|
|
self,
|
|
hidden_states: torch.FloatTensor,
|
|
attention_mask: torch.FloatTensor | None = None,
|
|
encoder_hidden_states: torch.FloatTensor | None = None,
|
|
encoder_attention_mask: torch.FloatTensor | None = None,
|
|
timestep: torch.LongTensor | None = None,
|
|
cross_attention_kwargs: Dict[str, Any] = None,
|
|
class_labels: torch.LongTensor | None = None,
|
|
_: Dict[str, torch.Tensor] | None = None,
|
|
) -> torch.FloatTensor:
|
|
assert isinstance(self.idx, int)
|
|
ref_data = self.ref_data
|
|
assert isinstance(ref_data, ReferenceData)
|
|
bank = self.bank
|
|
assert isinstance(bank, list)
|
|
|
|
uc_mask = ref_data.uc_mask
|
|
|
|
if self.use_ada_layer_norm:
|
|
norm_hidden_states = self.norm1(hidden_states, timestep)
|
|
elif self.use_ada_layer_norm_zero:
|
|
(
|
|
norm_hidden_states,
|
|
gate_msa,
|
|
shift_mlp,
|
|
scale_mlp,
|
|
gate_mlp,
|
|
) = self.norm1(
|
|
hidden_states,
|
|
timestep,
|
|
class_labels,
|
|
hidden_dtype=hidden_states.dtype,
|
|
)
|
|
else:
|
|
norm_hidden_states = self.norm1(hidden_states)
|
|
|
|
# 1. Retrieve lora scale.
|
|
lora_scale = (
|
|
cross_attention_kwargs.get("scale", 1.0)
|
|
if cross_attention_kwargs is not None
|
|
else 1.0
|
|
)
|
|
|
|
# 2. Prepare GLIGEN inputs
|
|
cross_attention_kwargs = (
|
|
cross_attention_kwargs.copy() if cross_attention_kwargs is not None else {}
|
|
)
|
|
gligen_kwargs = cross_attention_kwargs.pop("gligen", None)
|
|
|
|
# 1. Self-Attention
|
|
cross_attention_kwargs = (
|
|
cross_attention_kwargs if cross_attention_kwargs is not None else {}
|
|
)
|
|
|
|
if self.only_cross_attention:
|
|
attn_output = self.attn1(
|
|
norm_hidden_states,
|
|
encoder_hidden_states=(
|
|
encoder_hidden_states if self.only_cross_attention else None
|
|
),
|
|
attention_mask=attention_mask,
|
|
**cross_attention_kwargs,
|
|
)
|
|
else:
|
|
if ref_data.MODE == "write":
|
|
bank.append(norm_hidden_states.detach().clone())
|
|
attn_output = self.attn1(
|
|
norm_hidden_states,
|
|
encoder_hidden_states=(
|
|
encoder_hidden_states if self.only_cross_attention else None
|
|
),
|
|
attention_mask=attention_mask,
|
|
**cross_attention_kwargs,
|
|
)
|
|
if ref_data.MODE == "read":
|
|
style_fidelity = ref_data.style_fidelity
|
|
attention_auto_machine_weight = ref_data.attention_auto_machine_weight
|
|
do_classifier_free_guidance = ref_data.do_classifier_free_guidance
|
|
if attention_auto_machine_weight > self.attn_weight:
|
|
attn_output_uc = self.attn1(
|
|
norm_hidden_states,
|
|
encoder_hidden_states=torch.cat(
|
|
[norm_hidden_states] + self.bank, dim=1
|
|
),
|
|
# attention_mask=attention_mask,
|
|
**cross_attention_kwargs,
|
|
)
|
|
attn_output_c = attn_output_uc.clone()
|
|
if do_classifier_free_guidance and style_fidelity > 0:
|
|
attn_output_c[uc_mask] = self.attn1(
|
|
norm_hidden_states[uc_mask],
|
|
encoder_hidden_states=norm_hidden_states[uc_mask],
|
|
**cross_attention_kwargs,
|
|
)
|
|
attn_output = (
|
|
style_fidelity * attn_output_c
|
|
+ (1.0 - style_fidelity) * attn_output_uc
|
|
)
|
|
bank.clear()
|
|
else:
|
|
# 原始的自注意力（无 reference only
|
|
attn_output = self.attn1(
|
|
norm_hidden_states,
|
|
encoder_hidden_states=(
|
|
encoder_hidden_states if self.only_cross_attention else None
|
|
),
|
|
attention_mask=attention_mask,
|
|
**cross_attention_kwargs,
|
|
)
|
|
|
|
if self.use_ada_layer_norm_zero:
|
|
attn_output = gate_msa.unsqueeze(1) * attn_output
|
|
|
|
hidden_states = attn_output + hidden_states
|
|
|
|
# 2.5 GLIGEN Control
|
|
if gligen_kwargs is not None:
|
|
hidden_states = self.fuser(hidden_states, gligen_kwargs["objs"])
|
|
# 2.5 ends
|
|
|
|
# 2. Cross-Attention
|
|
if self.attn2 is not None:
|
|
norm_hidden_states = (
|
|
self.norm2(hidden_states, timestep)
|
|
if self.use_ada_layer_norm
|
|
else self.norm2(hidden_states)
|
|
)
|
|
|
|
attn_output = self.attn2(
|
|
norm_hidden_states,
|
|
encoder_hidden_states=encoder_hidden_states,
|
|
attention_mask=encoder_attention_mask,
|
|
**cross_attention_kwargs,
|
|
)
|
|
hidden_states = attn_output + hidden_states
|
|
|
|
gligen_kwargs = cross_attention_kwargs.pop("gligen", None)
|
|
|
|
# 3. Feed-forward
|
|
norm_hidden_states = self.norm3(hidden_states)
|
|
|
|
if self.use_ada_layer_norm_zero:
|
|
norm_hidden_states = (
|
|
norm_hidden_states * (1 + scale_mlp[:, None]) + shift_mlp[:, None]
|
|
)
|
|
|
|
ff_output = self.ff(norm_hidden_states, scale=lora_scale)
|
|
|
|
if self.use_ada_layer_norm_zero:
|
|
ff_output = gate_mlp.unsqueeze(1) * ff_output
|
|
|
|
hidden_states = ff_output + hidden_states
|
|
|
|
return hidden_states
|
|
|
|
|
|
class CrossAttnDownBlock2DReferenceOnly(CrossAttnDownBlock2D):
|
|
@classmethod
|
|
def from_module(
|
|
cls, module: CrossAttnDownBlock2D
|
|
) -> "CrossAttnDownBlock2DReferenceOnly":
|
|
module.__class__ = cls
|
|
return module
|
|
|
|
def forward(
|
|
self,
|
|
hidden_states: torch.FloatTensor,
|
|
temb: torch.FloatTensor | None = None,
|
|
encoder_hidden_states: torch.FloatTensor | None = None,
|
|
attention_mask: torch.FloatTensor | None = None,
|
|
cross_attention_kwargs: Dict[str, Any] | None = None,
|
|
encoder_attention_mask: torch.FloatTensor | None = None,
|
|
additional_residuals: torch.FloatTensor | None = None,
|
|
) -> Tuple[torch.FloatTensor | Tuple[torch.FloatTensor]]:
|
|
MODE = self.ref_data.MODE
|
|
gn_auto_machine_weight = self.ref_data.gn_auto_machine_weight
|
|
do_classifier_free_guidance = self.ref_data.do_classifier_free_guidance
|
|
style_fidelity = self.ref_data.style_fidelity
|
|
uc_mask = self.ref_data.uc_mask
|
|
|
|
eps = 1e-6
|
|
# TODO(Patrick, William) - attention mask is not used
|
|
output_states = ()
|
|
|
|
for i, (resnet, attn) in enumerate(zip(self.resnets, self.attentions)):
|
|
hidden_states = resnet(hidden_states, temb)
|
|
hidden_states = attn(
|
|
hidden_states,
|
|
encoder_hidden_states=encoder_hidden_states,
|
|
cross_attention_kwargs=cross_attention_kwargs,
|
|
attention_mask=attention_mask,
|
|
encoder_attention_mask=encoder_attention_mask,
|
|
return_dict=False,
|
|
)[0]
|
|
if MODE == "write" and gn_auto_machine_weight >= self.gn_weight:
|
|
var, mean = torch.var_mean(
|
|
hidden_states, dim=(2, 3), keepdim=True, correction=0
|
|
)
|
|
self.mean_bank.append([mean])
|
|
self.var_bank.append([var])
|
|
if MODE == "read" and (len(self.mean_bank) > 0 and len(self.var_bank) > 0):
|
|
var, mean = torch.var_mean(
|
|
hidden_states, dim=(2, 3), keepdim=True, correction=0
|
|
)
|
|
std = torch.maximum(var, torch.zeros_like(var) + eps) ** 0.5
|
|
mean_acc = sum(self.mean_bank[i]) / float(len(self.mean_bank[i]))
|
|
var_acc = sum(self.var_bank[i]) / float(len(self.var_bank[i]))
|
|
std_acc = torch.maximum(var_acc, torch.zeros_like(var_acc) + eps) ** 0.5
|
|
hidden_states_uc = (((hidden_states - mean) / std) * std_acc) + mean_acc
|
|
hidden_states_c = hidden_states_uc.clone()
|
|
if do_classifier_free_guidance and style_fidelity > 0:
|
|
hidden_states_c[uc_mask] = hidden_states[uc_mask]
|
|
hidden_states = (
|
|
style_fidelity * hidden_states_c
|
|
+ (1.0 - style_fidelity) * hidden_states_uc
|
|
)
|
|
|
|
output_states = output_states + (hidden_states,)
|
|
|
|
if MODE == "read":
|
|
self.mean_bank = []
|
|
self.var_bank = []
|
|
|
|
if self.downsamplers is not None:
|
|
for downsampler in self.downsamplers:
|
|
hidden_states = downsampler(hidden_states)
|
|
|
|
output_states = output_states + (hidden_states,)
|
|
|
|
return hidden_states, output_states
|
|
|
|
|
|
class DownBlock2DReferenceOnly(DownBlock2D):
|
|
@classmethod
|
|
def from_module(cls, module: DownBlock2D):
|
|
instance = cls()
|
|
instance.__dict__.update(module.__dict__)
|
|
return instance
|
|
|
|
def forward(
|
|
self,
|
|
hidden_states: torch.FloatTensor,
|
|
temb: torch.FloatTensor | None = None,
|
|
*args,
|
|
**kwargs,
|
|
) -> Tuple[torch.FloatTensor | Tuple[torch.FloatTensor]]:
|
|
|
|
MODE = self.ref_data.MODE
|
|
gn_auto_machine_weight = self.ref_data.gn_auto_machine_weight
|
|
do_classifier_free_guidance = self.ref_data.do_classifier_free_guidance
|
|
style_fidelity = self.ref_data.style_fidelity
|
|
uc_mask = self.ref_data.uc_mask
|
|
|
|
eps = 1e-6
|
|
output_states = ()
|
|
for i, resnet in enumerate(self.resnets):
|
|
hidden_states = resnet(hidden_states, temb)
|
|
if MODE == "write" and gn_auto_machine_weight >= self.gn_weight:
|
|
var, mean = torch.var_mean(
|
|
hidden_states, dim=(2, 3), keepdim=True, correction=0
|
|
)
|
|
self.mean_bank.append([mean])
|
|
self.var_bank.append([var])
|
|
if MODE == "read" and (len(self.mean_bank) > 0 and len(self.var_bank) > 0):
|
|
var, mean = torch.var_mean(
|
|
hidden_states, dim=(2, 3), keepdim=True, correction=0
|
|
)
|
|
std = torch.maximum(var, torch.zeros_like(var) + eps) ** 0.5
|
|
mean_acc = sum(self.mean_bank[i]) / float(len(self.mean_bank[i]))
|
|
var_acc = sum(self.var_bank[i]) / float(len(self.var_bank[i]))
|
|
std_acc = torch.maximum(var_acc, torch.zeros_like(var_acc) + eps) ** 0.5
|
|
hidden_states_uc = (((hidden_states - mean) / std) * std_acc) + mean_acc
|
|
hidden_states_c = hidden_states_uc.clone()
|
|
if do_classifier_free_guidance and style_fidelity > 0:
|
|
hidden_states_c[uc_mask] = hidden_states[uc_mask]
|
|
hidden_states = (
|
|
style_fidelity * hidden_states_c
|
|
+ (1.0 - style_fidelity) * hidden_states_uc
|
|
)
|
|
|
|
output_states = output_states + (hidden_states,)
|
|
|
|
if MODE == "read":
|
|
self.mean_bank = []
|
|
self.var_bank = []
|
|
|
|
if self.downsamplers is not None:
|
|
for downsampler in self.downsamplers:
|
|
hidden_states = downsampler(hidden_states)
|
|
|
|
output_states = output_states + (hidden_states,)
|
|
|
|
return hidden_states, output_states
|
|
|
|
|
|
class UNetMidBlock2DCrossAttnReferenceOnly(UNetMidBlock2DCrossAttn):
|
|
@classmethod
|
|
def from_module(cls, module: UNetMidBlock2DCrossAttn):
|
|
instance = cls()
|
|
instance.__dict__.update(module.__dict__)
|
|
return instance
|
|
|
|
def forward(
|
|
self,
|
|
hidden_states: torch.FloatTensor,
|
|
temb: torch.FloatTensor | None = None,
|
|
encoder_hidden_states: torch.FloatTensor | None = None,
|
|
attention_mask: torch.FloatTensor | None = None,
|
|
cross_attention_kwargs: Dict[str, Any] | None = None,
|
|
encoder_attention_mask: torch.FloatTensor | None = None,
|
|
) -> torch.FloatTensor:
|
|
return super().forward(
|
|
hidden_states,
|
|
temb,
|
|
encoder_hidden_states,
|
|
attention_mask,
|
|
cross_attention_kwargs,
|
|
encoder_attention_mask,
|
|
)
|
|
|
|
def forward(self, *args, **kwargs):
|
|
MODE = self.ref_data.MODE
|
|
gn_auto_machine_weight = self.ref_data.gn_auto_machine_weight
|
|
do_classifier_free_guidance = self.ref_data.do_classifier_free_guidance
|
|
style_fidelity = self.ref_data.style_fidelity
|
|
uc_mask = self.ref_data.uc_mask
|
|
eps = 1e-6
|
|
x = super().forward(*args, **kwargs)
|
|
if MODE == "write" and gn_auto_machine_weight >= self.gn_weight:
|
|
var, mean = torch.var_mean(x, dim=(2, 3), keepdim=True, correction=0)
|
|
self.mean_bank.append(mean)
|
|
self.var_bank.append(var)
|
|
if MODE == "read":
|
|
if len(self.mean_bank) > 0 and len(self.var_bank) > 0:
|
|
var, mean = torch.var_mean(x, dim=(2, 3), keepdim=True, correction=0)
|
|
std = torch.maximum(var, torch.zeros_like(var) + eps) ** 0.5
|
|
mean_acc = sum(self.mean_bank) / float(len(self.mean_bank))
|
|
var_acc = sum(self.var_bank) / float(len(self.var_bank))
|
|
std_acc = torch.maximum(var_acc, torch.zeros_like(var_acc) + eps) ** 0.5
|
|
x_uc = (((x - mean) / std) * std_acc) + mean_acc
|
|
x_c = x_uc.clone()
|
|
if do_classifier_free_guidance and style_fidelity > 0:
|
|
x_c[uc_mask] = x[uc_mask]
|
|
x = style_fidelity * x_c + (1.0 - style_fidelity) * x_uc
|
|
self.mean_bank = []
|
|
self.var_bank = []
|
|
return x
|
|
|
|
|
|
class UpBlock2DReferenceOnly(UpBlock2D):
|
|
@classmethod
|
|
def from_module(cls, module: UpBlock2D):
|
|
instance = cls()
|
|
instance.__dict__.update(module.__dict__)
|
|
return instance
|
|
|
|
def forward(
|
|
self,
|
|
hidden_states: torch.FloatTensor,
|
|
res_hidden_states_tuple: Tuple[torch.FloatTensor, ...],
|
|
temb: Optional[torch.FloatTensor] = None,
|
|
upsample_size: Optional[int] = None,
|
|
*args,
|
|
**kwargs,
|
|
) -> torch.FloatTensor:
|
|
MODE = self.ref_data.MODE
|
|
gn_auto_machine_weight = self.ref_data.gn_auto_machine_weight
|
|
do_classifier_free_guidance = self.ref_data.do_classifier_free_guidance
|
|
style_fidelity = self.ref_data.style_fidelity
|
|
uc_mask = self.ref_data.uc_mask
|
|
|
|
eps = 1e-6
|
|
for i, resnet in enumerate(self.resnets):
|
|
# pop res hidden states
|
|
res_hidden_states = res_hidden_states_tuple[-1]
|
|
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
|
hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
|
|
hidden_states = resnet(hidden_states, temb)
|
|
|
|
if MODE == "write" and gn_auto_machine_weight >= self.gn_weight:
|
|
var, mean = torch.var_mean(
|
|
hidden_states, dim=(2, 3), keepdim=True, correction=0
|
|
)
|
|
self.mean_bank.append([mean])
|
|
self.var_bank.append([var])
|
|
if MODE == "read" and (len(self.mean_bank) > 0 and len(self.var_bank) > 0):
|
|
var, mean = torch.var_mean(
|
|
hidden_states, dim=(2, 3), keepdim=True, correction=0
|
|
)
|
|
std = torch.maximum(var, torch.zeros_like(var) + eps) ** 0.5
|
|
mean_acc = sum(self.mean_bank[i]) / float(len(self.mean_bank[i]))
|
|
var_acc = sum(self.var_bank[i]) / float(len(self.var_bank[i]))
|
|
std_acc = torch.maximum(var_acc, torch.zeros_like(var_acc) + eps) ** 0.5
|
|
hidden_states_uc = (((hidden_states - mean) / std) * std_acc) + mean_acc
|
|
hidden_states_c = hidden_states_uc.clone()
|
|
if do_classifier_free_guidance and style_fidelity > 0:
|
|
hidden_states_c[uc_mask] = hidden_states[uc_mask]
|
|
hidden_states = (
|
|
style_fidelity * hidden_states_c
|
|
+ (1.0 - style_fidelity) * hidden_states_uc
|
|
)
|
|
|
|
if MODE == "read":
|
|
self.mean_bank = []
|
|
self.var_bank = []
|
|
|
|
if self.upsamplers is not None:
|
|
for upsampler in self.upsamplers:
|
|
hidden_states = upsampler(hidden_states, upsample_size)
|
|
|
|
return hidden_states
|
|
|
|
|
|
class CrossAttnUpBlock2DReferenceOnly(CrossAttnUpBlock2D):
|
|
@classmethod
|
|
def from_module(cls, module: CrossAttnUpBlock2D):
|
|
instance = cls()
|
|
instance.__dict__.update(module.__dict__)
|
|
return instance
|
|
|
|
def forward(
|
|
self,
|
|
hidden_states: torch.FloatTensor,
|
|
res_hidden_states_tuple: Tuple[torch.FloatTensor],
|
|
temb: torch.FloatTensor | None = None,
|
|
encoder_hidden_states: torch.FloatTensor | None = None,
|
|
cross_attention_kwargs: Dict[str, Any] | None = None,
|
|
upsample_size: int | None = None,
|
|
attention_mask: torch.FloatTensor | None = None,
|
|
encoder_attention_mask: torch.FloatTensor | None = None,
|
|
) -> torch.FloatTensor:
|
|
|
|
MODE = self.ref_data.MODE
|
|
gn_auto_machine_weight = self.ref_data.gn_auto_machine_weight
|
|
do_classifier_free_guidance = self.ref_data.do_classifier_free_guidance
|
|
style_fidelity = self.ref_data.style_fidelity
|
|
uc_mask = self.ref_data.uc_mask
|
|
|
|
eps = 1e-6
|
|
# TODO(Patrick, William) - attention mask is not used
|
|
for i, (resnet, attn) in enumerate(zip(self.resnets, self.attentions)):
|
|
# pop res hidden states
|
|
res_hidden_states = res_hidden_states_tuple[-1]
|
|
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
|
hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
|
|
hidden_states = resnet(hidden_states, temb)
|
|
hidden_states = attn(
|
|
hidden_states,
|
|
encoder_hidden_states=encoder_hidden_states,
|
|
cross_attention_kwargs=cross_attention_kwargs,
|
|
attention_mask=attention_mask,
|
|
encoder_attention_mask=encoder_attention_mask,
|
|
return_dict=False,
|
|
)[0]
|
|
|
|
if MODE == "write" and gn_auto_machine_weight >= self.gn_weight:
|
|
var, mean = torch.var_mean(
|
|
hidden_states, dim=(2, 3), keepdim=True, correction=0
|
|
)
|
|
self.mean_bank.append([mean])
|
|
self.var_bank.append([var])
|
|
if MODE == "read" and (len(self.mean_bank) > 0 and len(self.var_bank) > 0):
|
|
var, mean = torch.var_mean(
|
|
hidden_states, dim=(2, 3), keepdim=True, correction=0
|
|
)
|
|
std = torch.maximum(var, torch.zeros_like(var) + eps) ** 0.5
|
|
mean_acc = sum(self.mean_bank[i]) / float(len(self.mean_bank[i]))
|
|
var_acc = sum(self.var_bank[i]) / float(len(self.var_bank[i]))
|
|
std_acc = torch.maximum(var_acc, torch.zeros_like(var_acc) + eps) ** 0.5
|
|
hidden_states_uc = (((hidden_states - mean) / std) * std_acc) + mean_acc
|
|
hidden_states_c = hidden_states_uc.clone()
|
|
if do_classifier_free_guidance and style_fidelity > 0:
|
|
hidden_states_c[uc_mask] = hidden_states[uc_mask]
|
|
hidden_states = (
|
|
style_fidelity * hidden_states_c
|
|
+ (1.0 - style_fidelity) * hidden_states_uc
|
|
)
|
|
|
|
if MODE == "read":
|
|
self.mean_bank = []
|
|
self.var_bank = []
|
|
|
|
if self.upsamplers is not None:
|
|
for upsampler in self.upsamplers:
|
|
hidden_states = upsampler(hidden_states, upsample_size)
|
|
|
|
return hidden_states
|