Files

386 lines
19 KiB
Python

from functools import partial
import math
import torch
from torch import nn
from torch.nn import functional as F
class ResidualBlock(nn.Module):
def __init__(self, main, skip=None):
super().__init__()
self.main = nn.Sequential(*main)
self.skip = skip if skip else nn.Identity()
def forward(self, input):
return self.main(input) + self.skip(input)
class ResLinearBlock(ResidualBlock):
def __init__(self, f_in, f_mid, f_out, is_last=False):
skip = None if f_in == f_out else nn.Linear(f_in, f_out, bias=False)
super().__init__(
[
nn.Linear(f_in, f_mid),
nn.ReLU(inplace=True),
nn.Linear(f_mid, f_out),
nn.ReLU(inplace=True) if not is_last else nn.Identity(),
],
skip,
)
class Modulation2d(nn.Module):
def __init__(self, state, feats_in, c_out):
super().__init__()
self.state = state
self.layer = nn.Linear(feats_in, c_out * 2, bias=False)
def forward(self, input):
scales, shifts = self.layer(self.state["cond"]).chunk(2, dim=-1)
return torch.addcmul(
shifts[..., None, None], input, scales[..., None, None] + 1
)
class ResModConvBlock(ResidualBlock):
def __init__(self, state, feats_in, c_in, c_mid, c_out, is_last=False):
skip = None if c_in == c_out else nn.Conv2d(c_in, c_out, 1, bias=False)
super().__init__(
[
nn.Conv2d(c_in, c_mid, 3, padding=1),
nn.GroupNorm(1, c_mid, affine=False),
Modulation2d(state, feats_in, c_mid),
nn.ReLU(inplace=True),
nn.Conv2d(c_mid, c_out, 3, padding=1),
nn.GroupNorm(1, c_out, affine=False)
if not is_last
else nn.Identity(),
Modulation2d(state, feats_in, c_out)
if not is_last
else nn.Identity(),
nn.ReLU(inplace=True) if not is_last else nn.Identity(),
],
skip,
)
class SkipBlock(nn.Module):
def __init__(self, main, skip=None):
super().__init__()
self.main = nn.Sequential(*main)
self.skip = skip if skip else nn.Identity()
def forward(self, input):
return torch.cat([self.main(input), self.skip(input)], dim=1)
class FourierFeatures(nn.Module):
def __init__(self, in_features, out_features, std=1.0):
super().__init__()
assert out_features % 2 == 0
self.weight = nn.Parameter(
torch.randn([out_features // 2, in_features]) * std
)
self.weight.requires_grad_(False)
# self.register_buffer('weight', torch.randn([out_features // 2, in_features]) * std)
def forward(self, input):
f = 2 * math.pi * input @ self.weight.T
return torch.cat([f.cos(), f.sin()], dim=-1)
class SelfAttention2d(nn.Module):
def __init__(self, c_in, n_head=1, dropout_rate=0.1):
super().__init__()
assert c_in % n_head == 0
self.norm = nn.GroupNorm(1, c_in)
self.n_head = n_head
self.qkv_proj = nn.Conv2d(c_in, c_in * 3, 1)
self.out_proj = nn.Conv2d(c_in, c_in, 1)
self.dropout = (
nn.Identity()
) # nn.Dropout2d(dropout_rate, inplace=True)
def forward(self, input):
n, c, h, w = input.shape
qkv = self.qkv_proj(self.norm(input))
qkv = qkv.view(
[n, self.n_head * 3, c // self.n_head, h * w]
).transpose(2, 3)
q, k, v = qkv.chunk(3, dim=1)
scale = k.shape[3] ** -0.25
att = ((q * scale) @ (k.transpose(2, 3) * scale)).softmax(3)
y = (att @ v).transpose(2, 3).contiguous().view([n, c, h, w])
return input + self.dropout(self.out_proj(y))
def expand_to_planes(input, shape):
return input[..., None, None].repeat([1, 1, shape[2], shape[3]])
class CC12M1Model(nn.Module):
def __init__(self):
super().__init__()
self.shape = (3, 256, 256)
self.clip_model = "ViT-B/16"
self.min_t = 0.0
self.max_t = 1.0
c = 128 # The base channel count
cs = [c, c * 2, c * 2, c * 4, c * 4, c * 8, c * 8]
self.mapping_timestep_embed = FourierFeatures(1, 128)
self.mapping = nn.Sequential(
ResLinearBlock(512 + 128, 1024, 1024),
ResLinearBlock(1024, 1024, 1024, is_last=True),
)
with torch.no_grad():
for param in self.mapping.parameters():
param *= 0.5**0.5
self.state = {}
conv_block = partial(ResModConvBlock, self.state, 1024)
self.timestep_embed = FourierFeatures(1, 16)
self.down = nn.AvgPool2d(2)
self.up = nn.Upsample(
scale_factor=2, mode="bilinear", align_corners=False
)
self.net = nn.Sequential( # 256x256
conv_block(3 + 16, cs[0], cs[0]),
conv_block(cs[0], cs[0], cs[0]),
conv_block(cs[0], cs[0], cs[0]),
conv_block(cs[0], cs[0], cs[0]),
SkipBlock(
[
self.down, # 128x128
conv_block(cs[0], cs[1], cs[1]),
conv_block(cs[1], cs[1], cs[1]),
conv_block(cs[1], cs[1], cs[1]),
conv_block(cs[1], cs[1], cs[1]),
SkipBlock(
[
self.down, # 64x64
conv_block(cs[1], cs[2], cs[2]),
conv_block(cs[2], cs[2], cs[2]),
conv_block(cs[2], cs[2], cs[2]),
conv_block(cs[2], cs[2], cs[2]),
SkipBlock(
[
self.down, # 32x32
conv_block(cs[2], cs[3], cs[3]),
conv_block(cs[3], cs[3], cs[3]),
conv_block(cs[3], cs[3], cs[3]),
conv_block(cs[3], cs[3], cs[3]),
SkipBlock(
[
self.down, # 16x16
conv_block(cs[3], cs[4], cs[4]),
SelfAttention2d(
cs[4], cs[4] // 64
),
conv_block(cs[4], cs[4], cs[4]),
SelfAttention2d(
cs[4], cs[4] // 64
),
conv_block(cs[4], cs[4], cs[4]),
SelfAttention2d(
cs[4], cs[4] // 64
),
conv_block(cs[4], cs[4], cs[4]),
SelfAttention2d(
cs[4], cs[4] // 64
),
SkipBlock(
[
self.down, # 8x8
conv_block(
cs[4], cs[5], cs[5]
),
SelfAttention2d(
cs[5], cs[5] // 64
),
conv_block(
cs[5], cs[5], cs[5]
),
SelfAttention2d(
cs[5], cs[5] // 64
),
conv_block(
cs[5], cs[5], cs[5]
),
SelfAttention2d(
cs[5], cs[5] // 64
),
conv_block(
cs[5], cs[5], cs[5]
),
SelfAttention2d(
cs[5], cs[5] // 64
),
SkipBlock(
[
self.down, # 4x4
conv_block(
cs[5],
cs[6],
cs[6],
),
SelfAttention2d(
cs[6],
cs[6] // 64,
),
conv_block(
cs[6],
cs[6],
cs[6],
),
SelfAttention2d(
cs[6],
cs[6] // 64,
),
conv_block(
cs[6],
cs[6],
cs[6],
),
SelfAttention2d(
cs[6],
cs[6] // 64,
),
conv_block(
cs[6],
cs[6],
cs[6],
),
SelfAttention2d(
cs[6],
cs[6] // 64,
),
conv_block(
cs[6],
cs[6],
cs[6],
),
SelfAttention2d(
cs[6],
cs[6] // 64,
),
conv_block(
cs[6],
cs[6],
cs[6],
),
SelfAttention2d(
cs[6],
cs[6] // 64,
),
conv_block(
cs[6],
cs[6],
cs[6],
),
SelfAttention2d(
cs[6],
cs[6] // 64,
),
conv_block(
cs[6],
cs[6],
cs[5],
),
SelfAttention2d(
cs[5],
cs[5] // 64,
),
self.up,
]
),
conv_block(
cs[5] * 2, cs[5], cs[5]
),
SelfAttention2d(
cs[5], cs[5] // 64
),
conv_block(
cs[5], cs[5], cs[5]
),
SelfAttention2d(
cs[5], cs[5] // 64
),
conv_block(
cs[5], cs[5], cs[5]
),
SelfAttention2d(
cs[5], cs[5] // 64
),
conv_block(
cs[5], cs[5], cs[4]
),
SelfAttention2d(
cs[4], cs[4] // 64
),
self.up,
]
),
conv_block(
cs[4] * 2, cs[4], cs[4]
),
SelfAttention2d(
cs[4], cs[4] // 64
),
conv_block(cs[4], cs[4], cs[4]),
SelfAttention2d(
cs[4], cs[4] // 64
),
conv_block(cs[4], cs[4], cs[4]),
SelfAttention2d(
cs[4], cs[4] // 64
),
conv_block(cs[4], cs[4], cs[3]),
SelfAttention2d(
cs[3], cs[3] // 64
),
self.up,
]
),
conv_block(cs[3] * 2, cs[3], cs[3]),
conv_block(cs[3], cs[3], cs[3]),
conv_block(cs[3], cs[3], cs[3]),
conv_block(cs[3], cs[3], cs[2]),
self.up,
]
),
conv_block(cs[2] * 2, cs[2], cs[2]),
conv_block(cs[2], cs[2], cs[2]),
conv_block(cs[2], cs[2], cs[2]),
conv_block(cs[2], cs[2], cs[1]),
self.up,
]
),
conv_block(cs[1] * 2, cs[1], cs[1]),
conv_block(cs[1], cs[1], cs[1]),
conv_block(cs[1], cs[1], cs[1]),
conv_block(cs[1], cs[1], cs[0]),
self.up,
]
),
conv_block(cs[0] * 2, cs[0], cs[0]),
conv_block(cs[0], cs[0], cs[0]),
conv_block(cs[0], cs[0], cs[0]),
conv_block(cs[0], cs[0], 3, is_last=True),
)
with torch.no_grad():
for param in self.net.parameters():
param *= 0.5**0.5
def forward(self, input, timestep_embed, selfcond):
self.state["cond"] = selfcond
out = self.net(torch.cat([input, timestep_embed], dim=1))
self.state.clear()
return out