File size: 6,635 Bytes
aa8012e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import math

import torch
import torch.nn as nn


# FFN
def FeedForward(dim, mult=4):
    inner_dim = int(dim * mult)
    return nn.Sequential(
        nn.LayerNorm(dim),
        nn.Linear(dim, inner_dim, bias=False),
        nn.GELU(),
        nn.Linear(inner_dim, dim, bias=False),
    )


def reshape_tensor(x, heads):
    bs, length, width = x.shape
    # (bs, length, width) --> (bs, length, n_heads, dim_per_head)
    x = x.view(bs, length, heads, -1)
    # (bs, length, n_heads, dim_per_head) --> (bs, n_heads, length, dim_per_head)
    x = x.transpose(1, 2)
    # (bs, n_heads, length, dim_per_head) --> (bs*n_heads, length, dim_per_head)
    x = x.reshape(bs, heads, length, -1)
    return x


class PerceiverAttentionCA(nn.Module):
    def __init__(self, *, dim=3072, dim_head=128, heads=16, kv_dim=2048):
        super().__init__()
        self.scale = dim_head ** -0.5
        self.dim_head = dim_head
        self.heads = heads
        inner_dim = dim_head * heads

        self.norm1 = nn.LayerNorm(dim if kv_dim is None else kv_dim)
        self.norm2 = nn.LayerNorm(dim)

        self.to_q = nn.Linear(dim, inner_dim, bias=False)
        self.to_kv = nn.Linear(dim if kv_dim is None else kv_dim, inner_dim * 2, bias=False)
        self.to_out = nn.Linear(inner_dim, dim, bias=False)

    def forward(self, x, latents):
        """
        Args:
            x (torch.Tensor): image features
                shape (b, n1, D)
            latent (torch.Tensor): latent features
                shape (b, n2, D)
        """
        x = self.norm1(x)
        latents = self.norm2(latents)

        b, seq_len, _ = latents.shape

        q = self.to_q(latents)
        k, v = self.to_kv(x).chunk(2, dim=-1)

        q = reshape_tensor(q, self.heads)
        k = reshape_tensor(k, self.heads)
        v = reshape_tensor(v, self.heads)

        # attention
        scale = 1 / math.sqrt(math.sqrt(self.dim_head))
        weight = (q * scale) @ (k * scale).transpose(-2, -1)  # More stable with f16 than dividing afterwards
        weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
        out = weight @ v

        out = out.permute(0, 2, 1, 3).reshape(b, seq_len, -1)

        return self.to_out(out)


class PerceiverAttention(nn.Module):
    def __init__(self, *, dim, dim_head=64, heads=8, kv_dim=None):
        super().__init__()
        self.scale = dim_head ** -0.5
        self.dim_head = dim_head
        self.heads = heads
        inner_dim = dim_head * heads

        self.norm1 = nn.LayerNorm(dim if kv_dim is None else kv_dim)
        self.norm2 = nn.LayerNorm(dim)

        self.to_q = nn.Linear(dim, inner_dim, bias=False)
        self.to_kv = nn.Linear(dim if kv_dim is None else kv_dim, inner_dim * 2, bias=False)
        self.to_out = nn.Linear(inner_dim, dim, bias=False)

    def forward(self, x, latents):
        """
        Args:
            x (torch.Tensor): image features
                shape (b, n1, D)
            latent (torch.Tensor): latent features
                shape (b, n2, D)
        """
        x = self.norm1(x)
        latents = self.norm2(latents)

        b, seq_len, _ = latents.shape

        q = self.to_q(latents)
        kv_input = torch.cat((x, latents), dim=-2)
        k, v = self.to_kv(kv_input).chunk(2, dim=-1)

        q = reshape_tensor(q, self.heads)
        k = reshape_tensor(k, self.heads)
        v = reshape_tensor(v, self.heads)

        # attention
        scale = 1 / math.sqrt(math.sqrt(self.dim_head))
        weight = (q * scale) @ (k * scale).transpose(-2, -1)  # More stable with f16 than dividing afterwards
        weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
        out = weight @ v

        out = out.permute(0, 2, 1, 3).reshape(b, seq_len, -1)

        return self.to_out(out)


class IDFormer(nn.Module):
    """
    - perceiver resampler like arch (compared with previous MLP-like arch)
    - we concat id embedding (generated by arcface) and query tokens as latents
    - latents will attend each other and interact with vit features through cross-attention
    - vit features are multi-scaled and inserted into IDFormer in order, currently, each scale corresponds to two
      IDFormer layers
    """
    def __init__(
            self,
            dim=1024,
            depth=10,
            dim_head=64,
            heads=16,
            num_id_token=5,
            num_queries=32,
            output_dim=2048,
            ff_mult=4,
    ):
        super().__init__()

        self.num_id_token = num_id_token
        self.dim = dim
        self.num_queries = num_queries
        assert depth % 5 == 0
        self.depth = depth // 5
        scale = dim ** -0.5

        self.latents = nn.Parameter(torch.randn(1, num_queries, dim) * scale)
        self.proj_out = nn.Parameter(scale * torch.randn(dim, output_dim))

        self.layers = nn.ModuleList([])
        for _ in range(depth):
            self.layers.append(
                nn.ModuleList(
                    [
                        PerceiverAttention(dim=dim, dim_head=dim_head, heads=heads),
                        FeedForward(dim=dim, mult=ff_mult),
                    ]
                )
            )

        for i in range(5):
            setattr(
                self,
                f'mapping_{i}',
                nn.Sequential(
                    nn.Linear(1024, 1024),
                    nn.LayerNorm(1024),
                    nn.LeakyReLU(),
                    nn.Linear(1024, 1024),
                    nn.LayerNorm(1024),
                    nn.LeakyReLU(),
                    nn.Linear(1024, dim),
                ),
            )

        self.id_embedding_mapping = nn.Sequential(
            nn.Linear(1280, 1024),
            nn.LayerNorm(1024),
            nn.LeakyReLU(),
            nn.Linear(1024, 1024),
            nn.LayerNorm(1024),
            nn.LeakyReLU(),
            nn.Linear(1024, dim * num_id_token),
        )

    def forward(self, x, y):

        latents = self.latents.repeat(x.size(0), 1, 1)

        x = self.id_embedding_mapping(x)
        x = x.reshape(-1, self.num_id_token, self.dim)

        latents = torch.cat((latents, x), dim=1)

        for i in range(5):
            vit_feature = getattr(self, f'mapping_{i}')(y[i])
            ctx_feature = torch.cat((x, vit_feature), dim=1)
            for attn, ff in self.layers[i * self.depth: (i + 1) * self.depth]:
                latents = attn(ctx_feature, latents) + latents
                latents = ff(latents) + latents

        latents = latents[:, :self.num_queries]
        latents = latents @ self.proj_out
        return latents