File size: 15,068 Bytes
2f85de4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# python 3.7
"""Contains Super-Resolution Module described in EG3D."""

import numpy as np
import torch
from .official_stylegan2_model_helper import Conv2dLayer
from .official_stylegan2_model_helper import ToRGBLayer
from .official_stylegan2_model_helper import SynthesisLayer
from .official_stylegan2_model_helper import SynthesisBlock
from third_party.stylegan2_official_ops import upfirdn2d
from utils import eg3d_misc as misc

#----------------------------------------------------------------------------

# for 512x512 generation
class SuperresolutionHybrid8X(torch.nn.Module):
    def __init__(self, channels, img_resolution, sr_num_fp16_res, sr_antialias,
                num_fp16_res=4, conv_clamp=None, channel_base=None, channel_max=None,# IGNORE
                **block_kwargs):
        super().__init__()
        assert img_resolution == 512

        use_fp16 = sr_num_fp16_res > 0
        self.input_resolution = 128
        self.sr_antialias = sr_antialias
        self.block0 = SynthesisBlock(channels, 128, w_dim=512, resolution=256,
                img_channels=3, is_last=False, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)
        self.block1 = SynthesisBlock(128, 64, w_dim=512, resolution=512,
                img_channels=3, is_last=True, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)
        self.register_buffer('resample_filter', upfirdn2d.setup_filter([1,3,3,1]))

    def forward(self, rgb, x, ws, **block_kwargs):
        ws = ws[:, -1:, :].repeat(1, 3, 1)

        if x.shape[-1] != self.input_resolution:
            x = torch.nn.functional.interpolate(x, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)
            rgb = torch.nn.functional.interpolate(rgb, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)

        x, rgb = self.block0(x, rgb, ws, **block_kwargs)
        x, rgb = self.block1(x, rgb, ws, **block_kwargs)
        return rgb

#----------------------------------------------------------------------------

# for 256x256 generation
class SuperresolutionHybrid4X(torch.nn.Module):
    def __init__(self, channels, img_resolution, sr_num_fp16_res, sr_antialias,
                num_fp16_res=4, conv_clamp=None, channel_base=None, channel_max=None,# IGNORE
                **block_kwargs):
        super().__init__()
        assert img_resolution == 256
        use_fp16 = sr_num_fp16_res > 0
        self.sr_antialias = sr_antialias
        self.input_resolution = 128
        self.block0 = SynthesisBlockNoUp(channels, 128, w_dim=512, resolution=128,
                img_channels=3, is_last=False, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)
        self.block1 = SynthesisBlock(128, 64, w_dim=512, resolution=256,
                img_channels=3, is_last=True, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)
        self.register_buffer('resample_filter', upfirdn2d.setup_filter([1,3,3,1]))

    def forward(self, rgb, x, ws, **block_kwargs):
        ws = ws[:, -1:, :].repeat(1, 3, 1)

        if x.shape[-1] < self.input_resolution:
            x = torch.nn.functional.interpolate(x, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)
            rgb = torch.nn.functional.interpolate(rgb, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)

        x, rgb = self.block0(x, rgb, ws, **block_kwargs)
        x, rgb = self.block1(x, rgb, ws, **block_kwargs)
        return rgb

class SuperresolutionHybrid4X_conststyle(SuperresolutionHybrid4X):
    def forward(self, rgb, x, **block_kwargs):
        ws = torch.ones([x.shape[0], 3, 512]).float().to(x.device)
        if x.shape[-1] < self.input_resolution:
            x = torch.nn.functional.interpolate(x, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)
            rgb = torch.nn.functional.interpolate(rgb, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)

        x, rgb = self.block0(x, rgb, ws, **block_kwargs)
        x, rgb = self.block1(x, rgb, ws, **block_kwargs)
        return rgb

#----------------------------------------------------------------------------

# for 128 x 128 generation
class SuperresolutionHybrid2X(torch.nn.Module):
    def __init__(self, channels, img_resolution, sr_num_fp16_res, sr_antialias,
                num_fp16_res=4, conv_clamp=None, channel_base=None, channel_max=None,# IGNORE
                **block_kwargs):
        super().__init__()
        assert img_resolution == 128

        use_fp16 = sr_num_fp16_res > 0
        self.input_resolution = 64
        self.sr_antialias = sr_antialias
        self.block0 = SynthesisBlockNoUp(channels, 128, w_dim=512, resolution=64,
                img_channels=3, is_last=False, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)
        self.block1 = SynthesisBlock(128, 64, w_dim=512, resolution=128,
                img_channels=3, is_last=True, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)
        self.register_buffer('resample_filter', upfirdn2d.setup_filter([1,3,3,1]))

    def forward(self, rgb, x, ws, **block_kwargs):
        ws = ws[:, -1:, :].repeat(1, 3, 1)

        if x.shape[-1] != self.input_resolution:
            x = torch.nn.functional.interpolate(x, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)
            rgb = torch.nn.functional.interpolate(rgb, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)

        x, rgb = self.block0(x, rgb, ws, **block_kwargs)
        x, rgb = self.block1(x, rgb, ws, **block_kwargs)
        return rgb

#----------------------------------------------------------------------------

# TODO: Delete (here for backwards compatibility with old 256x256 models)
class SuperresolutionHybridDeepfp32(torch.nn.Module):
    def __init__(self, channels, img_resolution, sr_num_fp16_res,
                num_fp16_res=4, conv_clamp=None, channel_base=None, channel_max=None,# IGNORE
                **block_kwargs):
        super().__init__()
        assert img_resolution == 256
        use_fp16 = sr_num_fp16_res > 0

        self.input_resolution = 128
        self.block0 = SynthesisBlockNoUp(channels, 128, w_dim=512, resolution=128,
                img_channels=3, is_last=False, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)
        self.block1 = SynthesisBlock(128, 64, w_dim=512, resolution=256,
                img_channels=3, is_last=True, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)
        self.register_buffer('resample_filter', upfirdn2d.setup_filter([1,3,3,1]))

    def forward(self, rgb, x, ws, **block_kwargs):
        ws = ws[:, -1:, :].repeat(1, 3, 1)

        if x.shape[-1] < self.input_resolution:
            x = torch.nn.functional.interpolate(x, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)
            rgb = torch.nn.functional.interpolate(rgb, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)

        x, rgb = self.block0(x, rgb, ws, **block_kwargs)
        x, rgb = self.block1(x, rgb, ws, **block_kwargs)
        return rgb

#----------------------------------------------------------------------------

class SynthesisBlockNoUp(torch.nn.Module):
    def __init__(self,
        in_channels,                            # Number of input channels, 0 = first block.
        out_channels,                           # Number of output channels.
        w_dim,                                  # Intermediate latent (W) dimensionality.
        resolution,                             # Resolution of this block.
        img_channels,                           # Number of output color channels.
        is_last,                                # Is this the last block?
        architecture            = 'skip',       # Architecture: 'orig', 'skip', 'resnet'.
        resample_filter         = [1,3,3,1],    # Low-pass filter to apply when resampling activations.
        conv_clamp              = 256,          # Clamp the output of convolution layers to +-X, None = disable clamping.
        use_fp16                = False,        # Use FP16 for this block?
        fp16_channels_last      = False,        # Use channels-last memory format with FP16?
        fused_modconv_default   = True,         # Default value of fused_modconv. 'inference_only' = True for inference, False for training.
        **layer_kwargs,                         # Arguments for SynthesisLayer.
    ):
        assert architecture in ['orig', 'skip', 'resnet']
        super().__init__()
        self.in_channels = in_channels
        self.w_dim = w_dim
        self.resolution = resolution
        self.img_channels = img_channels
        self.is_last = is_last
        self.architecture = architecture
        self.use_fp16 = use_fp16
        self.channels_last = (use_fp16 and fp16_channels_last)
        self.fused_modconv_default = fused_modconv_default
        self.register_buffer('resample_filter', upfirdn2d.setup_filter(resample_filter))
        self.num_conv = 0
        self.num_torgb = 0

        if in_channels == 0:
            self.const = torch.nn.Parameter(torch.randn([out_channels, resolution, resolution]))

        if in_channels != 0:
            self.conv0 = SynthesisLayer(in_channels, out_channels, w_dim=w_dim, resolution=resolution,
                conv_clamp=conv_clamp, channels_last=self.channels_last, **layer_kwargs)
            self.num_conv += 1

        self.conv1 = SynthesisLayer(out_channels, out_channels, w_dim=w_dim, resolution=resolution,
            conv_clamp=conv_clamp, channels_last=self.channels_last, **layer_kwargs)
        self.num_conv += 1

        if is_last or architecture == 'skip':
            self.torgb = ToRGBLayer(out_channels, img_channels, w_dim=w_dim,
                conv_clamp=conv_clamp, channels_last=self.channels_last)
            self.num_torgb += 1

        if in_channels != 0 and architecture == 'resnet':
            self.skip = Conv2dLayer(in_channels, out_channels, kernel_size=1, bias=False, up=2,
                resample_filter=resample_filter, channels_last=self.channels_last)

    def forward(self, x, img, ws, force_fp32=False, fused_modconv=None, update_emas=False, **layer_kwargs):
        _ = update_emas # unused
        misc.assert_shape(ws, [None, self.num_conv + self.num_torgb, self.w_dim])
        w_iter = iter(ws.unbind(dim=1))
        if ws.device.type != 'cuda':
            force_fp32 = True
        dtype = torch.float16 if self.use_fp16 and not force_fp32 else torch.float32
        memory_format = torch.channels_last if self.channels_last and not force_fp32 else torch.contiguous_format
        if fused_modconv is None:
            fused_modconv = self.fused_modconv_default
        if fused_modconv == 'inference_only':
            fused_modconv = (not self.training)

        # Input.
        if self.in_channels == 0:
            x = self.const.to(dtype=dtype, memory_format=memory_format)
            x = x.unsqueeze(0).repeat([ws.shape[0], 1, 1, 1])
        else:
            misc.assert_shape(x, [None, self.in_channels, self.resolution, self.resolution])
            x = x.to(dtype=dtype, memory_format=memory_format)

        # Main layers.
        if self.in_channels == 0:
            x = self.conv1(x, next(w_iter), fused_modconv=fused_modconv, **layer_kwargs)
        elif self.architecture == 'resnet':
            y = self.skip(x, gain=np.sqrt(0.5))
            x = self.conv0(x, next(w_iter), fused_modconv=fused_modconv, **layer_kwargs)
            x = self.conv1(x, next(w_iter), fused_modconv=fused_modconv, gain=np.sqrt(0.5), **layer_kwargs)
            x = y.add_(x)
        else:
            x = self.conv0(x, next(w_iter), fused_modconv=fused_modconv, **layer_kwargs)
            x = self.conv1(x, next(w_iter), fused_modconv=fused_modconv, **layer_kwargs)

        # ToRGB.
        # if img is not None:
            # misc.assert_shape(img, [None, self.img_channels, self.resolution // 2, self.resolution // 2])
            # img = upfirdn2d.upsample2d(img, self.resample_filter)
        if self.is_last or self.architecture == 'skip':
            y = self.torgb(x, next(w_iter), fused_modconv=fused_modconv)
            y = y.to(dtype=torch.float32, memory_format=torch.contiguous_format)
            img = img.add_(y) if img is not None else y

        assert x.dtype == dtype
        assert img is None or img.dtype == torch.float32
        return x, img

    def extra_repr(self):
        return f'resolution={self.resolution:d}, architecture={self.architecture:s}'


#----------------------------------------------------------------------------

# for 512x512 generation
class SuperresolutionHybrid8XDC(torch.nn.Module):
    def __init__(self, channels, img_resolution, sr_num_fp16_res, sr_antialias,
                num_fp16_res=4, conv_clamp=None, channel_base=None, channel_max=None,# IGNORE
                **block_kwargs):
        super().__init__()
        assert img_resolution == 512

        use_fp16 = sr_num_fp16_res > 0
        self.input_resolution = 128
        self.sr_antialias = sr_antialias
        self.block0 = SynthesisBlock(channels, 256, w_dim=512, resolution=256,
                img_channels=3, is_last=False, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)
        self.block1 = SynthesisBlock(256, 128, w_dim=512, resolution=512,
                img_channels=3, is_last=True, use_fp16=use_fp16, conv_clamp=(256 if use_fp16 else None), **block_kwargs)

    def forward(self, rgb, x, ws, **block_kwargs):
        ws = ws[:, -1:, :].repeat(1, 3, 1)

        if x.shape[-1] != self.input_resolution:
            x = torch.nn.functional.interpolate(x, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)
            rgb = torch.nn.functional.interpolate(rgb, size=(self.input_resolution, self.input_resolution),
                                                  mode='bilinear', align_corners=False)

        x, rgb = self.block0(x, rgb, ws, **block_kwargs)
        x, rgb = self.block1(x, rgb, ws, **block_kwargs)
        return rgb

#----------------------------------------------------------------------------