File size: 6,932 Bytes
c9cd3be
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
259
"""
AUTOENCODER WITH ARCHTECTURE FROM VERSION 2
"""
from typing import Tuple

import torch
import torch.nn as nn
import torch.nn.functional as F


@torch.jit.script
def swish(x):
    return x * torch.sigmoid(x)


def Normalize(in_channels):
    return nn.GroupNorm(
        num_groups=32,
        num_channels=in_channels,
        eps=1e-6,
        affine=True
    )


class Upsample(nn.Module):
    def __init__(self, in_channels):
        super().__init__()
        self.conv = nn.Conv3d(
            in_channels,
            in_channels,
            kernel_size=3,
            stride=1,
            padding=1
        )

    def forward(self, x):
        x = F.interpolate(x, scale_factor=2.0, mode="nearest")
        x = self.conv(x)
        return x


class Downsample(nn.Module):
    def __init__(self, in_channels):
        super().__init__()
        self.conv = nn.Conv3d(
            in_channels,
            in_channels,
            kernel_size=3,
            stride=2,
            padding=0
        )

    def forward(self, x):
        pad = (0, 1, 0, 1, 0, 1)
        x = nn.functional.pad(x, pad, mode="constant", value=0)
        x = self.conv(x)
        return x


class ResBlock(nn.Module):
    def __init__(self, in_channels, out_channels=None):
        super().__init__()
        self.in_channels = in_channels
        self.out_channels = in_channels if out_channels is None else out_channels
        self.norm1 = Normalize(in_channels)
        self.conv1 = nn.Conv3d(
            in_channels,
            out_channels,
            kernel_size=3,
            stride=1,
            padding=1
        )
        self.norm2 = Normalize(out_channels)
        self.conv2 = nn.Conv3d(
            out_channels,
            out_channels,
            kernel_size=3,
            stride=1,
            padding=1
        )

        if self.in_channels != self.out_channels:
            self.nin_shortcut = nn.Conv3d(
                in_channels,
                out_channels,
                kernel_size=1,
                stride=1,
                padding=0
            )

    def forward(self, x):
        h = x
        h = self.norm1(h)
        h = F.silu(h)
        h = self.conv1(h)

        h = self.norm2(h)
        h = F.silu(h)
        h = self.conv2(h)

        if self.in_channels != self.out_channels:
            x = self.nin_shortcut(x)

        return x + h


class Encoder(nn.Module):
    def __init__(
            self,
            in_channels: int,
            n_channels: int,
            z_channels: int,
            ch_mult: Tuple[int],
            num_res_blocks: int,
            resolution: Tuple[int],
            attn_resolutions: Tuple[int],
            **ignorekwargs,
    ) -> None:
        super().__init__()
        self.in_channels = in_channels
        self.n_channels = n_channels
        self.num_resolutions = len(ch_mult)
        self.num_res_blocks = num_res_blocks
        self.resolution = resolution
        self.attn_resolutions = attn_resolutions

        curr_res = resolution
        in_ch_mult = (1,) + tuple(ch_mult)

        blocks = []
        # initial convolution
        blocks.append(
            nn.Conv3d(
                in_channels,
                n_channels,
                kernel_size=3,
                stride=1,
                padding=1
            )
        )

        # residual and downsampling blocks, with attention on smaller res (16x16)
        for i in range(self.num_resolutions):
            block_in_ch = n_channels * in_ch_mult[i]
            block_out_ch = n_channels * ch_mult[i]
            for _ in range(self.num_res_blocks):
                blocks.append(ResBlock(block_in_ch, block_out_ch))
                block_in_ch = block_out_ch

            if i != self.num_resolutions - 1:
                blocks.append(Downsample(block_in_ch))
                curr_res = tuple(ti // 2 for ti in curr_res)

        # normalise and convert to latent size
        blocks.append(Normalize(block_in_ch))
        blocks.append(
            nn.Conv3d(
                block_in_ch,
                z_channels,
                kernel_size=3,
                stride=1,
                padding=1
            )
        )

        self.blocks = nn.ModuleList(blocks)

    def forward(self, x):
        for block in self.blocks:
            x = block(x)
        return x


class Decoder(nn.Module):
    def __init__(
            self,
            n_channels: int,
            z_channels: int,
            out_channels: int,
            ch_mult: Tuple[int],
            num_res_blocks: int,
            resolution: Tuple[int],
            attn_resolutions: Tuple[int],
            **ignorekwargs,
    ) -> None:
        super().__init__()
        self.n_channels = n_channels
        self.z_channels = z_channels
        self.out_channels = out_channels
        self.ch_mult = ch_mult
        self.num_resolutions = len(ch_mult)
        self.num_res_blocks = num_res_blocks
        self.resolution = resolution
        self.attn_resolutions = attn_resolutions

        block_in_ch = n_channels * self.ch_mult[-1]
        curr_res = tuple(ti // 2 ** (self.num_resolutions - 1) for ti in resolution)

        blocks = []
        # initial conv
        blocks.append(
            nn.Conv3d(
                z_channels,
                block_in_ch,
                kernel_size=3,
                stride=1,
                padding=1
            )
        )

        for i in reversed(range(self.num_resolutions)):
            block_out_ch = n_channels * self.ch_mult[i]

            for _ in range(self.num_res_blocks):
                blocks.append(ResBlock(block_in_ch, block_out_ch))
                block_in_ch = block_out_ch

            if i != 0:
                blocks.append(Upsample(block_in_ch))
                curr_res = tuple(ti * 2 for ti in curr_res)

        blocks.append(Normalize(block_in_ch))
        blocks.append(
            nn.Conv3d(
                block_in_ch,
                out_channels,
                kernel_size=3,
                stride=1,
                padding=1
            )
        )

        self.blocks = nn.ModuleList(blocks)

    def forward(self, x):
        for block in self.blocks:
            x = block(x)
        return x


class AutoencoderKL(nn.Module):
    def __init__(self, embed_dim: int, hparams) -> None:
        super().__init__()
        self.encoder = Encoder(**hparams)
        self.decoder = Decoder(**hparams)
        self.quant_conv_mu = torch.nn.Conv3d(hparams["z_channels"], embed_dim, 1)
        self.quant_conv_log_sigma = torch.nn.Conv3d(hparams["z_channels"], embed_dim, 1)
        self.post_quant_conv = torch.nn.Conv3d(embed_dim, hparams["z_channels"], 1)
        self.embed_dim = embed_dim

    def decode(self, z):
        z = self.post_quant_conv(z)
        dec = self.decoder(z)
        return dec

    def reconstruct_ldm_outputs(self, z):
        x_hat = self.decode(z)
        return x_hat