File size: 12,810 Bytes
abe114d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
"""model.py — Neural network architectures for NSGF/NSGF++.

Contains:
  - VelocityMLP: MLP for 2D velocity field matching
  - VelocityUNet: UNet for image velocity field matching (NSGF + NSF)
  - PhaseTransitionPredictor: CNN for predicting transition time t_ϕ(x)

Reference: arXiv:2401.14069, Appendix E.1, E.2
"""

import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from typing import List, Optional


class SinusoidalPosEmb(nn.Module):
    def __init__(self, dim: int):
        super().__init__()
        self.dim = dim

    def forward(self, t: torch.Tensor) -> torch.Tensor:
        device = t.device
        half_dim = self.dim // 2
        emb = math.log(10000.0) / (half_dim - 1)
        emb = torch.exp(torch.arange(half_dim, device=device, dtype=torch.float32) * -emb)
        emb = t.float().unsqueeze(-1) * emb.unsqueeze(0)
        emb = torch.cat([emb.sin(), emb.cos()], dim=-1)
        return emb


class VelocityMLP(nn.Module):
    """MLP velocity field for 2D experiments.
    Paper: 3 hidden layers, 256 hidden units, SiLU activation.
    """
    def __init__(self, input_dim: int = 2, hidden_dim: int = 256,
                 num_hidden_layers: int = 3, time_emb_dim: int = 64):
        super().__init__()
        self.time_emb = SinusoidalPosEmb(time_emb_dim)
        layers = []
        layers.append(nn.Linear(input_dim + time_emb_dim, hidden_dim))
        layers.append(nn.SiLU())
        for _ in range(num_hidden_layers - 1):
            layers.append(nn.Linear(hidden_dim, hidden_dim))
            layers.append(nn.SiLU())
        layers.append(nn.Linear(hidden_dim, input_dim))
        self.net = nn.Sequential(*layers)

    def forward(self, x: torch.Tensor, t: torch.Tensor) -> torch.Tensor:
        t_emb = self.time_emb(t)
        xt = torch.cat([x, t_emb], dim=-1)
        return self.net(xt)


class ResBlock(nn.Module):
    """Residual block with AdaGN timestep conditioning."""
    def __init__(self, channels: int, emb_dim: int, out_channels: Optional[int] = None,
                 dropout: float = 0.0, use_scale_shift_norm: bool = True):
        super().__init__()
        self.out_channels = out_channels or channels
        self.use_scale_shift_norm = use_scale_shift_norm
        self.norm1 = nn.GroupNorm(32, channels)
        self.conv1 = nn.Conv2d(channels, self.out_channels, 3, padding=1)
        self.time_proj = nn.Sequential(
            nn.SiLU(),
            nn.Linear(emb_dim, 2 * self.out_channels if use_scale_shift_norm else self.out_channels),
        )
        self.norm2 = nn.GroupNorm(32, self.out_channels)
        self.dropout = nn.Dropout(dropout)
        self.conv2 = nn.Conv2d(self.out_channels, self.out_channels, 3, padding=1)
        if channels != self.out_channels:
            self.skip = nn.Conv2d(channels, self.out_channels, 1)
        else:
            self.skip = nn.Identity()
        nn.init.zeros_(self.conv2.weight)
        nn.init.zeros_(self.conv2.bias)

    def forward(self, x: torch.Tensor, emb: torch.Tensor) -> torch.Tensor:
        h = self.norm1(x)
        h = F.silu(h)
        h = self.conv1(h)
        emb_out = self.time_proj(emb)[:, :, None, None]
        if self.use_scale_shift_norm:
            scale, shift = emb_out.chunk(2, dim=1)
            h = self.norm2(h) * (1 + scale) + shift
        else:
            h = self.norm2(h + emb_out)
        h = F.silu(h)
        h = self.dropout(h)
        h = self.conv2(h)
        return h + self.skip(x)


class AttentionBlock(nn.Module):
    def __init__(self, channels: int, num_heads: int = 1, num_head_channels: int = -1):
        super().__init__()
        if num_head_channels > 0:
            self.num_heads = channels // num_head_channels
        else:
            self.num_heads = num_heads
        self.norm = nn.GroupNorm(32, channels)
        self.qkv = nn.Conv1d(channels, channels * 3, 1)
        self.proj = nn.Conv1d(channels, channels, 1)
        nn.init.zeros_(self.proj.weight)
        nn.init.zeros_(self.proj.bias)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        B, C, H, W = x.shape
        h = self.norm(x).view(B, C, -1)
        qkv = self.qkv(h).reshape(B, 3, self.num_heads, C // self.num_heads, -1)
        q, k, v = qkv[:, 0], qkv[:, 1], qkv[:, 2]
        scale = (C // self.num_heads) ** -0.5
        attn = torch.einsum("bhcn,bhcm->bhnm", q, k) * scale
        attn = attn.softmax(dim=-1)
        out = torch.einsum("bhnm,bhcm->bhcn", attn, v)
        out = out.reshape(B, C, -1)
        out = self.proj(out).view(B, C, H, W)
        return x + out


class Downsample(nn.Module):
    def __init__(self, channels: int):
        super().__init__()
        self.conv = nn.Conv2d(channels, channels, 3, stride=2, padding=1)
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return self.conv(x)


class Upsample(nn.Module):
    def __init__(self, channels: int):
        super().__init__()
        self.conv = nn.Conv2d(channels, channels, 3, padding=1)
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = F.interpolate(x, scale_factor=2, mode="nearest")
        return self.conv(x)


class VelocityUNet(nn.Module):
    """UNet velocity field for image experiments (Dhariwal & Nichol 2021).

    MNIST:   channels=32,  depth=1, ch_mult=[1,2,2],   heads=1
    CIFAR-10: channels=128, depth=2, ch_mult=[1,2,2,2], heads=4, head_ch=64
    """
    def __init__(self, image_size: int = 32, in_channels: int = 3,
                 model_channels: int = 128, num_res_blocks: int = 2,
                 channel_mult: List[int] = [1, 2, 2, 2],
                 attention_resolutions: List[int] = [16],
                 num_heads: int = 4, num_head_channels: int = 64,
                 dropout: float = 0.0, use_scale_shift_norm: bool = True):
        super().__init__()
        self.image_size = image_size
        self.in_channels = in_channels
        self.model_channels = model_channels

        time_dim = model_channels * 4
        self.time_embed = nn.Sequential(
            SinusoidalPosEmb(model_channels),
            nn.Linear(model_channels, time_dim), nn.SiLU(),
            nn.Linear(time_dim, time_dim),
        )
        self.input_conv = nn.Conv2d(in_channels, model_channels, 3, padding=1)

        self.down_blocks = nn.ModuleList()
        self.down_attns = nn.ModuleList()
        self.downsamplers = nn.ModuleList()

        ch = model_channels
        ds = image_size
        input_block_channels = [ch]

        for level, mult in enumerate(channel_mult):
            out_ch = model_channels * mult
            for _ in range(num_res_blocks):
                block = ResBlock(ch, time_dim, out_ch, dropout, use_scale_shift_norm)
                self.down_blocks.append(block)
                if ds in attention_resolutions:
                    self.down_attns.append(AttentionBlock(out_ch, num_heads, num_head_channels))
                else:
                    self.down_attns.append(nn.Identity())
                ch = out_ch
                input_block_channels.append(ch)
            if level < len(channel_mult) - 1:
                self.downsamplers.append(Downsample(ch))
                ds //= 2
                input_block_channels.append(ch)
            else:
                self.downsamplers.append(nn.Identity())

        self.mid_block1 = ResBlock(ch, time_dim, ch, dropout, use_scale_shift_norm)
        self.mid_attn = AttentionBlock(ch, num_heads, num_head_channels)
        self.mid_block2 = ResBlock(ch, time_dim, ch, dropout, use_scale_shift_norm)

        self.up_blocks = nn.ModuleList()
        self.up_attns = nn.ModuleList()
        self.upsamplers = nn.ModuleList()

        for level in reversed(range(len(channel_mult))):
            mult = channel_mult[level]
            out_ch = model_channels * mult
            for i in range(num_res_blocks + 1):
                skip_ch = input_block_channels.pop()
                block = ResBlock(ch + skip_ch, time_dim, out_ch, dropout, use_scale_shift_norm)
                self.up_blocks.append(block)
                if ds in attention_resolutions:
                    self.up_attns.append(AttentionBlock(out_ch, num_heads, num_head_channels))
                else:
                    self.up_attns.append(nn.Identity())
                ch = out_ch
            if level > 0:
                self.upsamplers.append(Upsample(ch))
                ds *= 2
            else:
                self.upsamplers.append(nn.Identity())

        self.out_norm = nn.GroupNorm(32, ch)
        self.out_conv = nn.Conv2d(ch, in_channels, 3, padding=1)
        nn.init.zeros_(self.out_conv.weight)
        nn.init.zeros_(self.out_conv.bias)

    def forward(self, x: torch.Tensor, t: torch.Tensor) -> torch.Tensor:
        emb = self.time_embed(t * 1000.0)
        h = self.input_conv(x)
        skips = [h]

        block_idx = 0
        for level in range(len(self.downsamplers)):
            for _ in range(self._get_num_res_blocks()):
                if block_idx < len(self.down_blocks):
                    h = self.down_blocks[block_idx](h, emb)
                    h = self.down_attns[block_idx](h)
                    skips.append(h)
                    block_idx += 1
            if not isinstance(self.downsamplers[level], nn.Identity):
                h = self.downsamplers[level](h)
                skips.append(h)

        h = self.mid_block1(h, emb)
        h = self.mid_attn(h)
        h = self.mid_block2(h, emb)

        block_idx = 0
        for level in range(len(self.upsamplers)):
            for _ in range(self._get_num_res_blocks() + 1):
                if block_idx < len(self.up_blocks):
                    skip = skips.pop()
                    h = torch.cat([h, skip], dim=1)
                    h = self.up_blocks[block_idx](h, emb)
                    h = self.up_attns[block_idx](h)
                    block_idx += 1
            if not isinstance(self.upsamplers[level], nn.Identity):
                h = self.upsamplers[level](h)

        h = self.out_norm(h)
        h = F.silu(h)
        h = self.out_conv(h)
        return h

    def _get_num_res_blocks(self):
        total_down = len(self.down_blocks)
        num_levels = len(self.downsamplers)
        return total_down // num_levels


class PhaseTransitionPredictor(nn.Module):
    """CNN predicting phase-transition time t_ϕ(x) ∈ [0, 1].
    4 conv layers (32→64→128→256), 3x3, AvgPool2d, FC + Sigmoid.
    """
    def __init__(self, in_channels: int = 1, image_size: int = 28,
                 conv_channels: List[int] = [32, 64, 128, 256]):
        super().__init__()
        layers = []
        ch = in_channels
        for out_ch in conv_channels:
            layers.extend([
                nn.Conv2d(ch, out_ch, kernel_size=3, stride=1, padding=1),
                nn.ReLU(inplace=True),
                nn.AvgPool2d(kernel_size=2, stride=2),
            ])
            ch = out_ch
        self.conv = nn.Sequential(*layers)
        final_size = image_size
        for _ in conv_channels:
            final_size = final_size // 2
        self.fc_input_dim = conv_channels[-1] * final_size * final_size
        self.fc = nn.Linear(self.fc_input_dim, 1)
        self.sigmoid = nn.Sigmoid()

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        h = self.conv(x)
        h = h.view(h.size(0), -1)
        h = self.fc(h)
        return self.sigmoid(h).squeeze(-1)


# Factory functions
def create_velocity_model_2d(config: dict) -> VelocityMLP:
    model_cfg = config.get("model", {})
    return VelocityMLP(
        input_dim=model_cfg.get("input_dim", 2),
        hidden_dim=model_cfg.get("hidden_dim", 256),
        num_hidden_layers=model_cfg.get("num_hidden_layers", 3),
        time_emb_dim=model_cfg.get("time_emb_dim", 64),
    )

def create_velocity_unet(config: dict) -> VelocityUNet:
    unet_cfg = config.get("unet", {})
    return VelocityUNet(
        image_size=config.get("image_size", 32),
        in_channels=config.get("in_channels", 3),
        model_channels=unet_cfg.get("model_channels", 128),
        num_res_blocks=unet_cfg.get("num_res_blocks", 2),
        channel_mult=unet_cfg.get("channel_mult", [1, 2, 2, 2]),
        attention_resolutions=unet_cfg.get("attention_resolutions", [16]),
        num_heads=unet_cfg.get("num_heads", 4),
        num_head_channels=unet_cfg.get("num_head_channels", 64),
        dropout=unet_cfg.get("dropout", 0.0),
        use_scale_shift_norm=unet_cfg.get("use_scale_shift_norm", True),
    )

def create_phase_predictor(config: dict) -> PhaseTransitionPredictor:
    tp_cfg = config.get("time_predictor", {})
    return PhaseTransitionPredictor(
        in_channels=config.get("in_channels", 1),
        image_size=config.get("image_size", 28),
        conv_channels=tp_cfg.get("conv_channels", [32, 64, 128, 256]),
    )