File size: 5,806 Bytes
a220803 |
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 |
import torch.nn as nn
from .normalize import Normalize
from .conv import CausalConv3d
import torch
import numpy as np
from einops import rearrange
from .block import Block
from .ops import video_to_image
class LinearAttention(Block):
def __init__(self, dim, heads=4, dim_head=32):
super().__init__()
self.heads = heads
hidden_dim = dim_head * heads
self.to_qkv = nn.Conv2d(dim, hidden_dim * 3, 1, bias=False)
self.to_out = nn.Conv2d(hidden_dim, dim, 1)
def forward(self, x):
b, c, h, w = x.shape
qkv = self.to_qkv(x)
q, k, v = rearrange(
qkv, "b (qkv heads c) h w -> qkv b heads c (h w)", heads=self.heads, qkv=3
)
k = k.softmax(dim=-1)
context = torch.einsum("bhdn,bhen->bhde", k, v)
out = torch.einsum("bhde,bhdn->bhen", context, q)
out = rearrange(
out, "b heads c (h w) -> b (heads c) h w", heads=self.heads, h=h, w=w
)
return self.to_out(out)
class LinAttnBlock(LinearAttention):
"""to match AttnBlock usage"""
def __init__(self, in_channels):
super().__init__(dim=in_channels, heads=1, dim_head=in_channels)
class AttnBlock3D(Block):
def __init__(self, in_channels):
super().__init__()
self.in_channels = in_channels
self.norm = Normalize(in_channels)
self.q = CausalConv3d(in_channels, in_channels, kernel_size=1, stride=1)
self.k = CausalConv3d(in_channels, in_channels, kernel_size=1, stride=1)
self.v = CausalConv3d(in_channels, in_channels, kernel_size=1, stride=1)
self.proj_out = CausalConv3d(in_channels, in_channels, kernel_size=1, stride=1)
def forward(self, x):
h_ = x
h_ = self.norm(h_)
q = self.q(h_)
k = self.k(h_)
v = self.v(h_)
# compute attention
b, c, t, h, w = q.shape
q = q.reshape(b * t, c, h * w)
q = q.permute(0, 2, 1) # b,hw,c
k = k.reshape(b * t, c, h * w) # b,c,hw
w_ = torch.bmm(q, k) # b,hw,hw w[b,i,j]=sum_c q[b,i,c]k[b,c,j]
w_ = w_ * (int(c) ** (-0.5))
w_ = torch.nn.functional.softmax(w_, dim=2)
# attend to values
v = v.reshape(b * t, c, h * w)
w_ = w_.permute(0, 2, 1) # b,hw,hw (first hw of k, second of q)
h_ = torch.bmm(v, w_) # b, c,hw (hw of q) h_[b,c,j] = sum_i v[b,c,i] w_[b,i,j]
h_ = h_.reshape(b, c, t, h, w)
h_ = self.proj_out(h_)
return x + h_
class AttnBlock(Block):
def __init__(self, in_channels):
super().__init__()
self.in_channels = in_channels
self.norm = Normalize(in_channels)
self.q = torch.nn.Conv2d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)
self.k = torch.nn.Conv2d(
in_channels, in_channels, kernel_size=1, stride=1, padding=0
)
self.v = torch.nn.Conv2d(
in_channels, in_channels, kernel_size=1, stride=1, padding=0
)
self.proj_out = torch.nn.Conv2d(
in_channels, in_channels, kernel_size=1, stride=1, padding=0
)
@video_to_image
def forward(self, x):
h_ = x
h_ = self.norm(h_)
q = self.q(h_)
k = self.k(h_)
v = self.v(h_)
# compute attention
b, c, h, w = q.shape
q = q.reshape(b, c, h * w)
q = q.permute(0, 2, 1) # b,hw,c
k = k.reshape(b, c, h * w) # b,c,hw
w_ = torch.bmm(q, k) # b,hw,hw w[b,i,j]=sum_c q[b,i,c]k[b,c,j]
w_ = w_ * (int(c) ** (-0.5))
w_ = torch.nn.functional.softmax(w_, dim=2)
# attend to values
v = v.reshape(b, c, h * w)
w_ = w_.permute(0, 2, 1) # b,hw,hw (first hw of k, second of q)
h_ = torch.bmm(v, w_) # b, c,hw (hw of q) h_[b,c,j] = sum_i v[b,c,i] w_[b,i,j]
h_ = h_.reshape(b, c, h, w)
h_ = self.proj_out(h_)
return x + h_
class TemporalAttnBlock(Block):
def __init__(self, in_channels):
super().__init__()
self.in_channels = in_channels
self.norm = Normalize(in_channels)
self.q = torch.nn.Conv3d(in_channels, in_channels, kernel_size=1, stride=1, padding=0)
self.k = torch.nn.Conv3d(
in_channels, in_channels, kernel_size=1, stride=1, padding=0
)
self.v = torch.nn.Conv3d(
in_channels, in_channels, kernel_size=1, stride=1, padding=0
)
self.proj_out = torch.nn.Conv3d(
in_channels, in_channels, kernel_size=1, stride=1, padding=0
)
def forward(self, x):
h_ = x
h_ = self.norm(h_)
q = self.q(h_)
k = self.k(h_)
v = self.v(h_)
# compute attention
b, c, t, h, w = q.shape
q = rearrange(q, "b c t h w -> (b h w) t c")
k = rearrange(k, "b c t h w -> (b h w) c t")
v = rearrange(v, "b c t h w -> (b h w) c t")
w_ = torch.bmm(q, k)
w_ = w_ * (int(c) ** (-0.5))
w_ = torch.nn.functional.softmax(w_, dim=2)
# attend to values
w_ = w_.permute(0, 2, 1)
h_ = torch.bmm(v, w_)
h_ = rearrange(h_, "(b h w) c t -> b c t h w", h=h, w=w)
h_ = self.proj_out(h_)
return x + h_
def make_attn(in_channels, attn_type="vanilla"):
assert attn_type in ["vanilla", "linear", "none", "vanilla3D"], f"attn_type {attn_type} unknown"
print(f"making attention of type '{attn_type}' with {in_channels} in_channels")
print(attn_type)
if attn_type == "vanilla":
return AttnBlock(in_channels)
elif attn_type == "vanilla3D":
return AttnBlock3D(in_channels)
elif attn_type == "none":
return nn.Identity(in_channels)
else:
return LinAttnBlock(in_channels) |