Spaces:
Configuration error
Configuration error
Create utils.py
Browse files- module/ip_adapter/utils.py +248 -0
module/ip_adapter/utils.py
ADDED
|
@@ -0,0 +1,248 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from collections import namedtuple, OrderedDict
|
| 3 |
+
from safetensors import safe_open
|
| 4 |
+
from .attention_processor import init_attn_proc
|
| 5 |
+
from .ip_adapter import MultiIPAdapterImageProjection
|
| 6 |
+
from .resampler import Resampler
|
| 7 |
+
from transformers import (
|
| 8 |
+
AutoModel, AutoImageProcessor,
|
| 9 |
+
CLIPVisionModelWithProjection, CLIPImageProcessor)
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
def init_adapter_in_unet(
|
| 13 |
+
unet,
|
| 14 |
+
image_proj_model=None,
|
| 15 |
+
pretrained_model_path_or_dict=None,
|
| 16 |
+
adapter_tokens=64,
|
| 17 |
+
embedding_dim=None,
|
| 18 |
+
use_lcm=False,
|
| 19 |
+
use_adaln=True,
|
| 20 |
+
):
|
| 21 |
+
device = unet.device
|
| 22 |
+
dtype = unet.dtype
|
| 23 |
+
if image_proj_model is None:
|
| 24 |
+
assert embedding_dim is not None, "embedding_dim must be provided if image_proj_model is None."
|
| 25 |
+
image_proj_model = Resampler(
|
| 26 |
+
embedding_dim=embedding_dim,
|
| 27 |
+
output_dim=unet.config.cross_attention_dim,
|
| 28 |
+
num_queries=adapter_tokens,
|
| 29 |
+
)
|
| 30 |
+
if pretrained_model_path_or_dict is not None:
|
| 31 |
+
if not isinstance(pretrained_model_path_or_dict, dict):
|
| 32 |
+
if pretrained_model_path_or_dict.endswith(".safetensors"):
|
| 33 |
+
state_dict = {"image_proj": {}, "ip_adapter": {}}
|
| 34 |
+
with safe_open(pretrained_model_path_or_dict, framework="pt", device=unet.device) as f:
|
| 35 |
+
for key in f.keys():
|
| 36 |
+
if key.startswith("image_proj."):
|
| 37 |
+
state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key)
|
| 38 |
+
elif key.startswith("ip_adapter."):
|
| 39 |
+
state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key)
|
| 40 |
+
else:
|
| 41 |
+
state_dict = torch.load(pretrained_model_path_or_dict, map_location=unet.device)
|
| 42 |
+
else:
|
| 43 |
+
state_dict = pretrained_model_path_or_dict
|
| 44 |
+
keys = list(state_dict.keys())
|
| 45 |
+
if "image_proj" not in keys and "ip_adapter" not in keys:
|
| 46 |
+
state_dict = revise_state_dict(state_dict)
|
| 47 |
+
|
| 48 |
+
# Creat IP cross-attention in unet.
|
| 49 |
+
attn_procs = init_attn_proc(unet, adapter_tokens, use_lcm, use_adaln)
|
| 50 |
+
unet.set_attn_processor(attn_procs)
|
| 51 |
+
|
| 52 |
+
# Load pretrinaed model if needed.
|
| 53 |
+
if pretrained_model_path_or_dict is not None:
|
| 54 |
+
if "ip_adapter" in state_dict.keys():
|
| 55 |
+
adapter_modules = torch.nn.ModuleList(unet.attn_processors.values())
|
| 56 |
+
missing, unexpected = adapter_modules.load_state_dict(state_dict["ip_adapter"], strict=False)
|
| 57 |
+
for mk in missing:
|
| 58 |
+
if "ln" not in mk:
|
| 59 |
+
raise ValueError(f"Missing keys in adapter_modules: {missing}")
|
| 60 |
+
if "image_proj" in state_dict.keys():
|
| 61 |
+
image_proj_model.load_state_dict(state_dict["image_proj"])
|
| 62 |
+
|
| 63 |
+
# Load image projectors into iterable ModuleList.
|
| 64 |
+
image_projection_layers = []
|
| 65 |
+
image_projection_layers.append(image_proj_model)
|
| 66 |
+
unet.encoder_hid_proj = MultiIPAdapterImageProjection(image_projection_layers)
|
| 67 |
+
|
| 68 |
+
# Adjust unet config to handle addtional ip hidden states.
|
| 69 |
+
unet.config.encoder_hid_dim_type = "ip_image_proj"
|
| 70 |
+
unet.to(dtype=dtype, device=device)
|
| 71 |
+
|
| 72 |
+
|
| 73 |
+
def load_adapter_to_pipe(
|
| 74 |
+
pipe,
|
| 75 |
+
pretrained_model_path_or_dict,
|
| 76 |
+
image_encoder_or_path=None,
|
| 77 |
+
feature_extractor_or_path=None,
|
| 78 |
+
use_clip_encoder=False,
|
| 79 |
+
adapter_tokens=64,
|
| 80 |
+
use_lcm=False,
|
| 81 |
+
use_adaln=True,
|
| 82 |
+
):
|
| 83 |
+
|
| 84 |
+
if not isinstance(pretrained_model_path_or_dict, dict):
|
| 85 |
+
if pretrained_model_path_or_dict.endswith(".safetensors"):
|
| 86 |
+
state_dict = {"image_proj": {}, "ip_adapter": {}}
|
| 87 |
+
with safe_open(pretrained_model_path_or_dict, framework="pt", device=pipe.device) as f:
|
| 88 |
+
for key in f.keys():
|
| 89 |
+
if key.startswith("image_proj."):
|
| 90 |
+
state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key)
|
| 91 |
+
elif key.startswith("ip_adapter."):
|
| 92 |
+
state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key)
|
| 93 |
+
else:
|
| 94 |
+
state_dict = torch.load(pretrained_model_path_or_dict, map_location=pipe.device)
|
| 95 |
+
else:
|
| 96 |
+
state_dict = pretrained_model_path_or_dict
|
| 97 |
+
keys = list(state_dict.keys())
|
| 98 |
+
if "image_proj" not in keys and "ip_adapter" not in keys:
|
| 99 |
+
state_dict = revise_state_dict(state_dict)
|
| 100 |
+
|
| 101 |
+
# load CLIP image encoder here if it has not been registered to the pipeline yet
|
| 102 |
+
if image_encoder_or_path is not None:
|
| 103 |
+
if isinstance(image_encoder_or_path, str):
|
| 104 |
+
feature_extractor_or_path = image_encoder_or_path if feature_extractor_or_path is None else feature_extractor_or_path
|
| 105 |
+
|
| 106 |
+
image_encoder_or_path = (
|
| 107 |
+
CLIPVisionModelWithProjection.from_pretrained(
|
| 108 |
+
image_encoder_or_path
|
| 109 |
+
) if use_clip_encoder else
|
| 110 |
+
AutoModel.from_pretrained(image_encoder_or_path)
|
| 111 |
+
)
|
| 112 |
+
|
| 113 |
+
if feature_extractor_or_path is not None:
|
| 114 |
+
if isinstance(feature_extractor_or_path, str):
|
| 115 |
+
feature_extractor_or_path = (
|
| 116 |
+
CLIPImageProcessor() if use_clip_encoder else
|
| 117 |
+
AutoImageProcessor.from_pretrained(feature_extractor_or_path)
|
| 118 |
+
)
|
| 119 |
+
|
| 120 |
+
# create image encoder if it has not been registered to the pipeline yet
|
| 121 |
+
if hasattr(pipe, "image_encoder") and getattr(pipe, "image_encoder", None) is None:
|
| 122 |
+
image_encoder = image_encoder_or_path.to(pipe.device, dtype=pipe.dtype)
|
| 123 |
+
pipe.register_modules(image_encoder=image_encoder)
|
| 124 |
+
else:
|
| 125 |
+
image_encoder = pipe.image_encoder
|
| 126 |
+
|
| 127 |
+
# create feature extractor if it has not been registered to the pipeline yet
|
| 128 |
+
if hasattr(pipe, "feature_extractor") and getattr(pipe, "feature_extractor", None) is None:
|
| 129 |
+
feature_extractor = feature_extractor_or_path
|
| 130 |
+
pipe.register_modules(feature_extractor=feature_extractor)
|
| 131 |
+
else:
|
| 132 |
+
feature_extractor = pipe.feature_extractor
|
| 133 |
+
|
| 134 |
+
# load adapter into unet
|
| 135 |
+
unet = getattr(pipe, pipe.unet_name) if not hasattr(pipe, "unet") else pipe.unet
|
| 136 |
+
attn_procs = init_attn_proc(unet, adapter_tokens, use_lcm, use_adaln)
|
| 137 |
+
unet.set_attn_processor(attn_procs)
|
| 138 |
+
image_proj_model = Resampler(
|
| 139 |
+
embedding_dim=image_encoder.config.hidden_size,
|
| 140 |
+
output_dim=unet.config.cross_attention_dim,
|
| 141 |
+
num_queries=adapter_tokens,
|
| 142 |
+
)
|
| 143 |
+
|
| 144 |
+
# Load pretrinaed model if needed.
|
| 145 |
+
if "ip_adapter" in state_dict.keys():
|
| 146 |
+
adapter_modules = torch.nn.ModuleList(unet.attn_processors.values())
|
| 147 |
+
missing, unexpected = adapter_modules.load_state_dict(state_dict["ip_adapter"], strict=False)
|
| 148 |
+
for mk in missing:
|
| 149 |
+
if "ln" not in mk:
|
| 150 |
+
raise ValueError(f"Missing keys in adapter_modules: {missing}")
|
| 151 |
+
if "image_proj" in state_dict.keys():
|
| 152 |
+
image_proj_model.load_state_dict(state_dict["image_proj"])
|
| 153 |
+
|
| 154 |
+
# convert IP-Adapter Image Projection layers to diffusers
|
| 155 |
+
image_projection_layers = []
|
| 156 |
+
image_projection_layers.append(image_proj_model)
|
| 157 |
+
unet.encoder_hid_proj = MultiIPAdapterImageProjection(image_projection_layers)
|
| 158 |
+
|
| 159 |
+
# Adjust unet config to handle addtional ip hidden states.
|
| 160 |
+
unet.config.encoder_hid_dim_type = "ip_image_proj"
|
| 161 |
+
unet.to(dtype=pipe.dtype, device=pipe.device)
|
| 162 |
+
|
| 163 |
+
|
| 164 |
+
def revise_state_dict(old_state_dict_or_path, map_location="cpu"):
|
| 165 |
+
new_state_dict = OrderedDict()
|
| 166 |
+
new_state_dict["image_proj"] = OrderedDict()
|
| 167 |
+
new_state_dict["ip_adapter"] = OrderedDict()
|
| 168 |
+
if isinstance(old_state_dict_or_path, str):
|
| 169 |
+
old_state_dict = torch.load(old_state_dict_or_path, map_location=map_location)
|
| 170 |
+
else:
|
| 171 |
+
old_state_dict = old_state_dict_or_path
|
| 172 |
+
for name, weight in old_state_dict.items():
|
| 173 |
+
if name.startswith("image_proj_model."):
|
| 174 |
+
new_state_dict["image_proj"][name[len("image_proj_model."):]] = weight
|
| 175 |
+
elif name.startswith("adapter_modules."):
|
| 176 |
+
new_state_dict["ip_adapter"][name[len("adapter_modules."):]] = weight
|
| 177 |
+
return new_state_dict
|
| 178 |
+
|
| 179 |
+
|
| 180 |
+
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
|
| 181 |
+
def encode_image(image_encoder, feature_extractor, image, device, num_images_per_prompt, output_hidden_states=None):
|
| 182 |
+
dtype = next(image_encoder.parameters()).dtype
|
| 183 |
+
|
| 184 |
+
if not isinstance(image, torch.Tensor):
|
| 185 |
+
image = feature_extractor(image, return_tensors="pt").pixel_values
|
| 186 |
+
|
| 187 |
+
image = image.to(device=device, dtype=dtype)
|
| 188 |
+
if output_hidden_states:
|
| 189 |
+
image_enc_hidden_states = image_encoder(image, output_hidden_states=True).hidden_states[-2]
|
| 190 |
+
image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
|
| 191 |
+
return image_enc_hidden_states
|
| 192 |
+
else:
|
| 193 |
+
if isinstance(image_encoder, CLIPVisionModelWithProjection):
|
| 194 |
+
# CLIP image encoder.
|
| 195 |
+
image_embeds = image_encoder(image).image_embeds
|
| 196 |
+
else:
|
| 197 |
+
# DINO image encoder.
|
| 198 |
+
image_embeds = image_encoder(image).last_hidden_state
|
| 199 |
+
image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
|
| 200 |
+
return image_embeds
|
| 201 |
+
|
| 202 |
+
|
| 203 |
+
def prepare_training_image_embeds(
|
| 204 |
+
image_encoder, feature_extractor,
|
| 205 |
+
ip_adapter_image, ip_adapter_image_embeds,
|
| 206 |
+
device, drop_rate, output_hidden_state, idx_to_replace=None
|
| 207 |
+
):
|
| 208 |
+
if ip_adapter_image_embeds is None:
|
| 209 |
+
if not isinstance(ip_adapter_image, list):
|
| 210 |
+
ip_adapter_image = [ip_adapter_image]
|
| 211 |
+
|
| 212 |
+
# if len(ip_adapter_image) != len(unet.encoder_hid_proj.image_projection_layers):
|
| 213 |
+
# raise ValueError(
|
| 214 |
+
# f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
|
| 215 |
+
# )
|
| 216 |
+
|
| 217 |
+
image_embeds = []
|
| 218 |
+
for single_ip_adapter_image in ip_adapter_image:
|
| 219 |
+
if idx_to_replace is None:
|
| 220 |
+
idx_to_replace = torch.rand(len(single_ip_adapter_image)) < drop_rate
|
| 221 |
+
zero_ip_adapter_image = torch.zeros_like(single_ip_adapter_image)
|
| 222 |
+
single_ip_adapter_image[idx_to_replace] = zero_ip_adapter_image[idx_to_replace]
|
| 223 |
+
single_image_embeds = encode_image(
|
| 224 |
+
image_encoder, feature_extractor, single_ip_adapter_image, device, 1, output_hidden_state
|
| 225 |
+
)
|
| 226 |
+
single_image_embeds = torch.stack([single_image_embeds], dim=1) # FIXME
|
| 227 |
+
|
| 228 |
+
image_embeds.append(single_image_embeds)
|
| 229 |
+
else:
|
| 230 |
+
repeat_dims = [1]
|
| 231 |
+
image_embeds = []
|
| 232 |
+
for single_image_embeds in ip_adapter_image_embeds:
|
| 233 |
+
if do_classifier_free_guidance:
|
| 234 |
+
single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
|
| 235 |
+
single_image_embeds = single_image_embeds.repeat(
|
| 236 |
+
num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:]))
|
| 237 |
+
)
|
| 238 |
+
single_negative_image_embeds = single_negative_image_embeds.repeat(
|
| 239 |
+
num_images_per_prompt, *(repeat_dims * len(single_negative_image_embeds.shape[1:]))
|
| 240 |
+
)
|
| 241 |
+
single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds])
|
| 242 |
+
else:
|
| 243 |
+
single_image_embeds = single_image_embeds.repeat(
|
| 244 |
+
num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:]))
|
| 245 |
+
)
|
| 246 |
+
image_embeds.append(single_image_embeds)
|
| 247 |
+
|
| 248 |
+
return image_embeds
|