File size: 6,789 Bytes
3e8e153
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import json
import math
from pathlib import Path

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


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

    def forward(self, t):
        half = self.dim // 2
        freqs = torch.exp(
            -math.log(10000) * torch.arange(half, device=t.device) / max(half - 1, 1)
        )
        args = t[:, None] * freqs[None, :]
        emb = torch.cat([torch.sin(args), torch.cos(args)], dim=-1)
        if self.dim % 2 == 1:
            emb = F.pad(emb, (0, 1))
        return emb


class ResBlock(nn.Module):
    def __init__(self, in_channels, out_channels, time_dim):
        super().__init__()
        self.norm1 = nn.GroupNorm(8, in_channels)
        self.conv1 = nn.Conv2d(in_channels, out_channels, 3, padding=1)
        self.time = nn.Sequential(nn.SiLU(), nn.Linear(time_dim, out_channels))
        self.norm2 = nn.GroupNorm(8, out_channels)
        self.conv2 = nn.Conv2d(out_channels, out_channels, 3, padding=1)
        self.skip = (
            nn.Conv2d(in_channels, out_channels, 1)
            if in_channels != out_channels
            else nn.Identity()
        )

    def forward(self, x, temb):
        h = self.conv1(F.silu(self.norm1(x)))
        h = h + self.time(temb)[:, :, None, None]
        h = self.conv2(F.silu(self.norm2(h)))
        return h + self.skip(x)


class Downsample(nn.Module):
    def __init__(self, channels):
        super().__init__()
        self.conv = nn.Conv2d(channels, channels, 4, stride=2, padding=1)

    def forward(self, x):
        return self.conv(x)


class Upsample(nn.Module):
    def __init__(self, channels):
        super().__init__()
        self.conv = nn.Conv2d(channels, channels, 3, padding=1)

    def forward(self, x):
        x = F.interpolate(x, scale_factor=2, mode="nearest")
        return self.conv(x)


class FlowMatchingModelConfig:
    def __init__(self, in_channels=3, base_channels=64):
        self.in_channels = in_channels
        self.base_channels = base_channels

    def to_dict(self):
        return {
            "model_type": "flow_matching_unet",
            "in_channels": self.in_channels,
            "base_channels": self.base_channels,
            "architecture": "UNetVelocity",
        }

    @classmethod
    def from_dict(cls, data):
        return cls(
            in_channels=data.get("in_channels", 3),
            base_channels=data.get("base_channels", 64),
        )


class FlowMatchingModel(nn.Module):
    def __init__(self, config=None, in_channels=3, base_channels=64):
        super().__init__()
        if config is None:
            config = FlowMatchingModelConfig(
                in_channels=in_channels,
                base_channels=base_channels,
            )
        self.config = config

        time_dim = self.config.base_channels * 4
        self.time_mlp = nn.Sequential(
            SinusoidalTimeEmbedding(self.config.base_channels),
            nn.Linear(self.config.base_channels, time_dim),
            nn.SiLU(),
            nn.Linear(time_dim, time_dim),
        )

        c1 = self.config.base_channels
        c2 = self.config.base_channels * 2
        c3 = self.config.base_channels * 4
        c4 = self.config.base_channels * 4

        self.init_conv = nn.Conv2d(self.config.in_channels, c1, 3, padding=1)
        self.enc1 = nn.ModuleList(
            [ResBlock(c1, c1, time_dim), ResBlock(c1, c1, time_dim)]
        )
        self.down1 = Downsample(c1)
        self.enc2 = nn.ModuleList(
            [ResBlock(c1, c2, time_dim), ResBlock(c2, c2, time_dim)]
        )
        self.down2 = Downsample(c2)
        self.enc3 = nn.ModuleList(
            [ResBlock(c2, c3, time_dim), ResBlock(c3, c3, time_dim)]
        )
        self.down3 = Downsample(c3)

        self.mid1 = ResBlock(c3, c4, time_dim)
        self.mid2 = ResBlock(c4, c4, time_dim)

        self.up3 = Upsample(c4)
        self.dec3 = nn.ModuleList(
            [ResBlock(c4 + c3, c3, time_dim), ResBlock(c3, c3, time_dim)]
        )
        self.up2 = Upsample(c3)
        self.dec2 = nn.ModuleList(
            [ResBlock(c3 + c2, c2, time_dim), ResBlock(c2, c2, time_dim)]
        )
        self.up1 = Upsample(c2)
        self.dec1 = nn.ModuleList(
            [ResBlock(c2 + c1, c1, time_dim), ResBlock(c1, c1, time_dim)]
        )

        self.out = nn.Sequential(
            nn.GroupNorm(8, c1),
            nn.SiLU(),
            nn.Conv2d(c1, self.config.in_channels, 3, padding=1),
        )

    def run_blocks(self, blocks, x, temb):
        for block in blocks:
            x = block(x, temb)
        return x

    def forward(self, x, t):
        temb = self.time_mlp(t)
        h = self.init_conv(x)

        s1 = self.run_blocks(self.enc1, h, temb)
        h = self.down1(s1)
        s2 = self.run_blocks(self.enc2, h, temb)
        h = self.down2(s2)
        s3 = self.run_blocks(self.enc3, h, temb)
        h = self.down3(s3)

        h = self.mid2(self.mid1(h, temb), temb)

        h = self.up3(h)
        h = self.run_blocks(self.dec3, torch.cat([h, s3], dim=1), temb)
        h = self.up2(h)
        h = self.run_blocks(self.dec2, torch.cat([h, s2], dim=1), temb)
        h = self.up1(h)
        h = self.run_blocks(self.dec1, torch.cat([h, s1], dim=1), temb)

        return self.out(h)

    @classmethod
    def from_pretrained(cls, model_path):
        path = Path(model_path)
        if not path.exists():
            raise FileNotFoundError(f"Model path does not exist: {path}")

        config_path = path / "config.json"
        if not config_path.exists():
            raise FileNotFoundError(f"Missing config.json in {path}")

        with open(config_path, "r", encoding="utf-8") as f:
            data = json.load(f)

        config = FlowMatchingModelConfig.from_dict(data)
        model = cls(config=config)

        weight_file = path / "pytorch_model.bin"
        if not weight_file.exists():
            weight_file = path / "model_final.pt"
        if not weight_file.exists():
            raise FileNotFoundError(f"No weights found in {path}")

        state = torch.load(weight_file, map_location="cpu")
        if isinstance(state, dict) and "model" in state and isinstance(state["model"], dict):
            state = state["model"]
        model.load_state_dict(state)
        model.eval()
        return model

    def save_pretrained(self, save_directory):
        path = Path(save_directory)
        path.mkdir(parents=True, exist_ok=True)

        with open(path / "config.json", "w", encoding="utf-8") as f:
            json.dump(self.config.to_dict(), f, indent=2)

        torch.save(self.state_dict(), path / "pytorch_model.bin")
        return str(path)


UNetVelocity = FlowMatchingModel