Spaces:
Runtime error
Runtime error
File size: 10,360 Bytes
f85e212 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 |
from typing import Optional, Tuple, Union
import torch
import torch.nn as nn
import torch.utils.checkpoint
from .embeddings import TimeEmbbeding
from .unet_blocks import (
CrossAttnDownBlock2D,
CrossAttnUpBlock2D,
DownBlock2D,
UNetMidBlock2DCrossAttn,
UpBlock2D,
get_down_block,
get_up_block,
)
class TimestepEmbedding(nn.Module):
def __init__(self, channel, time_embed_dim, act_fn="silu"):
super().__init__()
self.linear_1 = nn.Linear(channel, time_embed_dim)
self.act = None
if act_fn == "silu":
self.act = nn.SiLU()
self.linear_2 = nn.Linear(time_embed_dim, time_embed_dim)
def forward(self, sample):
sample = self.linear_1(sample)
if self.act is not None:
sample = self.act(sample)
sample = self.linear_2(sample)
return sample
class UNet2DConditionModel(nn.Module):
r"""
UNet2DConditionModel is a conditional 2D UNet model that takes in a noisy sample, conditional state, and a timestep
and returns sample shaped output.
Parameters:
sample_size (`int`, *optional*): The size of the input sample.
in_channels (`int`, *optional*, defaults to 4): The number of channels in the input sample.
out_channels (`int`, *optional*, defaults to 4): The number of channels in the output.
center_input_sample (`bool`, *optional*, defaults to `False`): Whether to center the input sample.
flip_sin_to_cos (`bool`, *optional*, defaults to `False`):
Whether to flip the sin to cos in the time embedding.
freq_shift (`int`, *optional*, defaults to 0): The frequency shift to apply to the time embedding.
down_block_types (`Tuple[str]`, *optional*, defaults to `("CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "DownBlock2D")`):
The tuple of downsample blocks to use.
up_block_types (`Tuple[str]`, *optional*, defaults to `("UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D",)`):
The tuple of upsample blocks to use.
block_out_channels (`Tuple[int]`, *optional*, defaults to `(320, 640, 1280, 1280)`):
The tuple of output channels for each block.
layers_per_block (`int`, *optional*, defaults to 2): The number of layers per block.
downsample_padding (`int`, *optional*, defaults to 1): The padding to use for the downsampling convolution.
mid_block_scale_factor (`float`, *optional*, defaults to 1.0): The scale factor to use for the mid block.
act_fn (`str`, *optional*, defaults to `"silu"`): The activation function to use.
norm_num_groups (`int`, *optional*, defaults to 32): The number of groups to use for the normalization.
norm_eps (`float`, *optional*, defaults to 1e-5): The epsilon to use for the normalization.
cross_attention_dim (`int`, *optional*, defaults to 1280): The dimension of the cross attention features.
attention_head_dim (`int`, *optional*, defaults to 8): The dimension of the attention heads.
"""
_supports_gradient_checkpointing = True
def __init__(
self,
sample_size: Optional[int] = None,
in_channels: int = 4,
out_channels: int = 4,
center_input_sample: bool = False,
flip_sin_to_cos: bool = True,
freq_shift: int = 0,
down_block_types: Tuple[str] = (
"CrossAttnDownBlock2D",
"CrossAttnDownBlock2D",
"CrossAttnDownBlock2D",
"DownBlock2D",
),
up_block_types: Tuple[str] = ("UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D"),
block_out_channels: Tuple[int] = (320, 640, 1280, 1280),
layers_per_block: int = 2,
downsample_padding: int = 1,
mid_block_scale_factor: float = 1,
act_fn: str = "silu",
norm_num_groups: int = 32,
norm_eps: float = 1e-5,
cross_attention_dim: int = 768,
attention_head_dim: int = 8,
):
super().__init__()
self.sample_size = sample_size
time_embed_dim = block_out_channels[0] * 4
self.emb = nn.Embedding(2, cross_attention_dim)
# input
self.conv_in = nn.Conv2d(in_channels, block_out_channels[0], kernel_size=3, padding=(1, 1))
# time
self.time_embedding = TimeEmbbeding(block_out_channels[0], time_embed_dim)
self.down_blocks = nn.ModuleList([])
self.mid_block = None
self.up_blocks = nn.ModuleList([])
# down
output_channel = block_out_channels[0]
for i, down_block_type in enumerate(down_block_types):
input_channel = output_channel
output_channel = block_out_channels[i]
is_final_block = i == len(block_out_channels) - 1
down_block = get_down_block(
down_block_type,
num_layers=layers_per_block,
in_channels=input_channel,
out_channels=output_channel,
temb_channels=time_embed_dim,
add_downsample=not is_final_block,
resnet_eps=norm_eps,
resnet_act_fn=act_fn,
resnet_groups=norm_num_groups,
cross_attention_dim=cross_attention_dim,
attn_num_head_channels=attention_head_dim,
downsample_padding=downsample_padding,
)
self.down_blocks.append(down_block)
# mid
self.mid_block = UNetMidBlock2DCrossAttn(
in_channels=block_out_channels[-1],
temb_channels=time_embed_dim,
resnet_eps=norm_eps,
resnet_act_fn=act_fn,
output_scale_factor=mid_block_scale_factor,
resnet_time_scale_shift="default",
cross_attention_dim=cross_attention_dim,
attn_num_head_channels=attention_head_dim,
resnet_groups=norm_num_groups,
)
# up
reversed_block_out_channels = list(reversed(block_out_channels))
output_channel = reversed_block_out_channels[0]
for i, up_block_type in enumerate(up_block_types):
prev_output_channel = output_channel
output_channel = reversed_block_out_channels[i]
input_channel = reversed_block_out_channels[min(i + 1, len(block_out_channels) - 1)]
is_final_block = i == len(block_out_channels) - 1
up_block = get_up_block(
up_block_type,
num_layers=layers_per_block + 1,
in_channels=input_channel,
out_channels=output_channel,
prev_output_channel=prev_output_channel,
temb_channels=time_embed_dim,
add_upsample=not is_final_block,
resnet_eps=norm_eps,
resnet_act_fn=act_fn,
resnet_groups=norm_num_groups,
cross_attention_dim=cross_attention_dim,
attn_num_head_channels=attention_head_dim,
)
self.up_blocks.append(up_block)
prev_output_channel = output_channel
# out
self.conv_norm_out = nn.GroupNorm(num_channels=block_out_channels[0], num_groups=norm_num_groups, eps=norm_eps)
self.conv_act = nn.SiLU()
self.conv_out = nn.Conv2d(block_out_channels[0], out_channels, 3, padding=1)
def forward(
self,
sample: torch.FloatTensor,
t: torch.Tensor,
encoder_hidden_states: torch.Tensor = None,
self_cond: torch.Tensor = None
):
encoder_hidden_states = self.emb(encoder_hidden_states)
# encoder_hidden_states = None # ------------------------ WARNING Disabled ---------------------
"""r
Args:
sample (`torch.FloatTensor`): (batch, channel, height, width) noisy inputs tensor
timestep (`torch.FloatTensor` or `float` or `int`): (batch) timesteps
encoder_hidden_states (`torch.FloatTensor`): (batch, channel, height, width) encoder hidden states
Returns:
[`~models.unet_2d_condition.UNet2DConditionOutput`] or `tuple`:
[`~models.unet_2d_condition.UNet2DConditionOutput`] if `return_dict` is True, otherwise a `tuple`. When
returning a tuple, the first element is the sample tensor.
"""
# 0. center input if necessary
# if self.config.center_input_sample:
# sample = 2 * sample - 1.0
# 1. time
t_emb = self.time_embedding(t)
# 2. pre-process
sample = self.conv_in(sample)
# 3. down
down_block_res_samples = (sample,)
for downsample_block in self.down_blocks:
if hasattr(downsample_block, "attentions") and downsample_block.attentions is not None:
sample, res_samples = downsample_block(
hidden_states=sample,
temb=t_emb,
encoder_hidden_states=encoder_hidden_states,
)
else:
sample, res_samples = downsample_block(hidden_states=sample, temb=t_emb)
down_block_res_samples += res_samples
# 4. mid
sample = self.mid_block(sample, t_emb, encoder_hidden_states=encoder_hidden_states)
# 5. up
for upsample_block in self.up_blocks:
res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)]
if hasattr(upsample_block, "attentions") and upsample_block.attentions is not None:
sample = upsample_block(
hidden_states=sample,
temb=t_emb,
res_hidden_states_tuple=res_samples,
encoder_hidden_states=encoder_hidden_states,
)
else:
sample = upsample_block(hidden_states=sample, temb=t_emb, res_hidden_states_tuple=res_samples)
# 6. post-process
# make sure hidden states is in float32
# when running in half-precision
sample = self.conv_norm_out(sample.float()).type(sample.dtype)
sample = self.conv_act(sample)
sample = self.conv_out(sample)
return sample, []
|