forked from huggingface/diffusers
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathunet_2d_condition.py
More file actions
178 lines (146 loc) 路 6.44 KB
/
Copy pathunet_2d_condition.py
File metadata and controls
178 lines (146 loc) 路 6.44 KB
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
from typing import Dict, Union
import torch
import torch.nn as nn
from ..configuration_utils import ConfigMixin, register_to_config
from ..modeling_utils import ModelMixin
from .embeddings import TimestepEmbedding, Timesteps
from .unet_blocks import UNetMidBlock2DCrossAttn, get_down_block, get_up_block
class UNet2DConditionModel(ModelMixin, ConfigMixin):
@register_to_config
def __init__(
self,
sample_size=None,
in_channels=4,
out_channels=4,
center_input_sample=False,
flip_sin_to_cos=True,
freq_shift=0,
down_block_types=("CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "DownBlock2D"),
up_block_types=("UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D"),
block_out_channels=(320, 640, 1280, 1280),
layers_per_block=2,
downsample_padding=1,
mid_block_scale_factor=1,
act_fn="silu",
norm_num_groups=32,
norm_eps=1e-5,
attention_head_dim=8,
):
super().__init__()
self.sample_size = sample_size
time_embed_dim = block_out_channels[0] * 4
# input
self.conv_in = nn.Conv2d(in_channels, block_out_channels[0], kernel_size=3, padding=(1, 1))
# time
self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
timestep_input_dim = block_out_channels[0]
self.time_embedding = TimestepEmbedding(timestep_input_dim, 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,
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",
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,
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,
timestep: Union[torch.Tensor, float, int],
encoder_hidden_states: torch.Tensor,
) -> Dict[str, torch.FloatTensor]:
# 0. center input if necessary
if self.config.center_input_sample:
sample = 2 * sample - 1.0
# 1. time
timesteps = timestep
if not torch.is_tensor(timesteps):
timesteps = torch.tensor([timesteps], dtype=torch.long, device=sample.device)
elif torch.is_tensor(timesteps) and len(timesteps.shape) == 0:
timesteps = timesteps[None].to(sample.device)
t_emb = self.time_proj(timesteps)
emb = self.time_embedding(t_emb)
# 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=emb, encoder_hidden_states=encoder_hidden_states
)
else:
sample, res_samples = downsample_block(hidden_states=sample, temb=emb)
down_block_res_samples += res_samples
# 4. mid
sample = self.mid_block(sample, 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=emb,
res_hidden_states_tuple=res_samples,
encoder_hidden_states=encoder_hidden_states,
)
else:
sample = upsample_block(hidden_states=sample, temb=emb, res_hidden_states_tuple=res_samples)
# 6. post-process
sample = self.conv_norm_out(sample)
sample = self.conv_act(sample)
sample = self.conv_out(sample)
output = {"sample": sample}
return output