File size: 15,956 Bytes
713ec7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
bf85ddb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
713ec7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dd97a63
 
 
 
 
 
 
 
 
 
713ec7d
 
 
dd97a63
 
 
 
 
 
 
 
bf85ddb
713ec7d
dd97a63
 
 
 
 
 
 
bf85ddb
713ec7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ed16a12
 
 
 
 
 
 
 
bf85ddb
713ec7d
ed16a12
 
 
 
 
 
 
bf85ddb
713ec7d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
329
330
331
332
333
334
335
336
337
338
339
340
import torch
import torch.fft as fft
from diffusers.models.unet_2d_condition import logger
from diffusers.utils import is_torch_version
from typing import Any, Dict, List, Optional, Tuple, Union


def isinstance_str(x: object, cls_name: str):
    """
    Checks whether x has any class *named* cls_name in its ancestry.
    Doesn't require access to the class's implementation.
    
    Useful for patching!
    """

    for _cls in x.__class__.__mro__:
        if _cls.__name__ == cls_name:
            return True
    
    return False


def Fourier_filter(x, threshold, scale):
    dtype = x.dtype
    x = x.type(torch.float32)
    # FFT
    x_freq = fft.fftn(x, dim=(-2, -1))
    x_freq = fft.fftshift(x_freq, dim=(-2, -1))
    
    B, C, H, W = x_freq.shape
    mask = torch.ones((B, C, H, W)).cuda() 

    crow, ccol = H // 2, W //2
    mask[..., crow - threshold:crow + threshold, ccol - threshold:ccol + threshold] = scale
    x_freq = x_freq * mask

    # IFFT
    x_freq = fft.ifftshift(x_freq, dim=(-2, -1))
    x_filtered = fft.ifftn(x_freq, dim=(-2, -1)).real
    
    x_filtered = x_filtered.type(dtype)
    return x_filtered


def register_upblock2d(model):
    def up_forward(self):
        def forward(hidden_states, res_hidden_states_tuple, temb=None, upsample_size=None):
            for resnet in self.resnets:
                # pop res hidden states
                res_hidden_states = res_hidden_states_tuple[-1]
                res_hidden_states_tuple = res_hidden_states_tuple[:-1]
                #print(f"in upblock2d, hidden states shape: {hidden_states.shape}")
                hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)

                if self.training and self.gradient_checkpointing:

                    def create_custom_forward(module):
                        def custom_forward(*inputs):
                            return module(*inputs)

                        return custom_forward

                    if is_torch_version(">=", "1.11.0"):
                        hidden_states = torch.utils.checkpoint.checkpoint(
                            create_custom_forward(resnet), hidden_states, temb, use_reentrant=False
                        )
                    else:
                        hidden_states = torch.utils.checkpoint.checkpoint(
                            create_custom_forward(resnet), hidden_states, temb
                        )
                else:
                    hidden_states = resnet(hidden_states, temb)

            if self.upsamplers is not None:
                for upsampler in self.upsamplers:
                    hidden_states = upsampler(hidden_states, upsample_size)

            return hidden_states
        
        return forward
    
    for i, upsample_block in enumerate(model.unet.up_blocks):
        if isinstance_str(upsample_block, "UpBlock2D"):
            upsample_block.forward = up_forward(upsample_block)


def register_free_upblock2d(model, b1=1.2, b2=1.4, s1=0.9, s2=0.2):
    def up_forward(self):
        def forward(hidden_states, res_hidden_states_tuple, temb=None, upsample_size=None):
            for resnet in self.resnets:
                # pop res hidden states
                res_hidden_states = res_hidden_states_tuple[-1]
                res_hidden_states_tuple = res_hidden_states_tuple[:-1]
                #print(f"in free upblock2d, hidden states shape: {hidden_states.shape}")
                
                # # --------------- FreeU code -----------------------
                # # Only operate on the first two stages
                # if hidden_states.shape[1] == 1280:
                #     hidden_states[:,:640] = hidden_states[:,:640] * self.b1
                #     res_hidden_states = Fourier_filter(res_hidden_states, threshold=1, scale=self.s1)
                # if hidden_states.shape[1] == 640:
                #     hidden_states[:,:320] = hidden_states[:,:320] * self.b2
                #     res_hidden_states = Fourier_filter(res_hidden_states, threshold=1, scale=self.s2)
                # # ---------------------------------------------------------

                # --------------- FreeU code -----------------------
                # Only operate on the first two stages
                if hidden_states.shape[1] == 1280:
                    hidden_mean = hidden_states.mean(1).unsqueeze(1)
                    B = hidden_mean.shape[0]
                    hidden_max, _ = torch.max(hidden_mean.view(B, -1), dim=-1, keepdim=True) 
                    hidden_min, _ = torch.min(hidden_mean.view(B, -1), dim=-1, keepdim=True)

                    hidden_mean = (hidden_mean - hidden_min.unsqueeze(2).unsqueeze(3)) / (hidden_max - hidden_min).unsqueeze(2).unsqueeze(3)
                    
                    hidden_states[:,:640] = hidden_states[:,:640] * ((self.b1 - 1 ) * hidden_mean + 1)
                    res_hidden_states = Fourier_filter(res_hidden_states, threshold=1, scale=self.s1)
                if hidden_states.shape[1] == 640:
                    hidden_mean = hidden_states.mean(1).unsqueeze(1)
                    B = hidden_mean.shape[0]
                    hidden_max, _ = torch.max(hidden_mean.view(B, -1), dim=-1, keepdim=True) 
                    hidden_min, _ = torch.min(hidden_mean.view(B, -1), dim=-1, keepdim=True)
                    hidden_mean = (hidden_mean - hidden_min.unsqueeze(2).unsqueeze(3)) / (hidden_max - hidden_min).unsqueeze(2).unsqueeze(3)
                    
                    hidden_states[:,:320] = hidden_states[:,:320] * ((self.b2 - 1 ) * hidden_mean + 1)
                    res_hidden_states = Fourier_filter(res_hidden_states, threshold=1, scale=self.s2)
                # ---------------------------------------------------------

                hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)

                if self.training and self.gradient_checkpointing:

                    def create_custom_forward(module):
                        def custom_forward(*inputs):
                            return module(*inputs)

                        return custom_forward

                    if is_torch_version(">=", "1.11.0"):
                        hidden_states = torch.utils.checkpoint.checkpoint(
                            create_custom_forward(resnet), hidden_states, temb, use_reentrant=False
                        )
                    else:
                        hidden_states = torch.utils.checkpoint.checkpoint(
                            create_custom_forward(resnet), hidden_states, temb
                        )
                else:
                    hidden_states = resnet(hidden_states, temb)

            if self.upsamplers is not None:
                for upsampler in self.upsamplers:
                    hidden_states = upsampler(hidden_states, upsample_size)

            return hidden_states
        
        return forward
    
    for i, upsample_block in enumerate(model.unet.up_blocks):
        if isinstance_str(upsample_block, "UpBlock2D"):
            upsample_block.forward = up_forward(upsample_block)
            setattr(upsample_block, 'b1', b1)
            setattr(upsample_block, 'b2', b2)
            setattr(upsample_block, 's1', s1)
            setattr(upsample_block, 's2', s2)


def register_crossattn_upblock2d(model):
    def up_forward(self):
        def forward(
            hidden_states: torch.FloatTensor,
            res_hidden_states_tuple: Tuple[torch.FloatTensor, ...],
            temb: Optional[torch.FloatTensor] = None,
            encoder_hidden_states: Optional[torch.FloatTensor] = None,
            cross_attention_kwargs: Optional[Dict[str, Any]] = None,
            upsample_size: Optional[int] = None,
            attention_mask: Optional[torch.FloatTensor] = None,
            encoder_attention_mask: Optional[torch.FloatTensor] = None,
        ):
            for resnet, attn in zip(self.resnets, self.attentions):
                # pop res hidden states
                #print(f"in crossatten upblock2d, hidden states shape: {hidden_states.shape}")
                res_hidden_states = res_hidden_states_tuple[-1]
                res_hidden_states_tuple = res_hidden_states_tuple[:-1]
                hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)

                if self.training and self.gradient_checkpointing:

                    def create_custom_forward(module, return_dict=None):
                        def custom_forward(*inputs):
                            if return_dict is not None:
                                return module(*inputs, return_dict=return_dict)
                            else:
                                return module(*inputs)

                        return custom_forward

                    ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
                    hidden_states = torch.utils.checkpoint.checkpoint(
                        create_custom_forward(resnet),
                        hidden_states,
                        temb,
                        **ckpt_kwargs,
                    )
                    hidden_states = torch.utils.checkpoint.checkpoint(
                        create_custom_forward(attn, return_dict=False),
                        hidden_states,
                        encoder_hidden_states,
                        None,  # timestep
                        None,  # class_labels
                        cross_attention_kwargs,
                        attention_mask,
                        encoder_attention_mask,
                        **ckpt_kwargs,
                    )[0]
                else:
                    hidden_states = resnet(hidden_states, temb)
                    hidden_states = attn(
                        hidden_states,
                        encoder_hidden_states=encoder_hidden_states,
                        cross_attention_kwargs=cross_attention_kwargs,
                        attention_mask=attention_mask,
                        encoder_attention_mask=encoder_attention_mask,
                        return_dict=False,
                    )[0]

            if self.upsamplers is not None:
                for upsampler in self.upsamplers:
                    hidden_states = upsampler(hidden_states, upsample_size)

            return hidden_states
        
        return forward
    
    for i, upsample_block in enumerate(model.unet.up_blocks):
        if isinstance_str(upsample_block, "CrossAttnUpBlock2D"):
            upsample_block.forward = up_forward(upsample_block)


def register_free_crossattn_upblock2d(model, b1=1.2, b2=1.4, s1=0.9, s2=0.2):
    def up_forward(self):
        def forward(
            hidden_states: torch.FloatTensor,
            res_hidden_states_tuple: Tuple[torch.FloatTensor, ...],
            temb: Optional[torch.FloatTensor] = None,
            encoder_hidden_states: Optional[torch.FloatTensor] = None,
            cross_attention_kwargs: Optional[Dict[str, Any]] = None,
            upsample_size: Optional[int] = None,
            attention_mask: Optional[torch.FloatTensor] = None,
            encoder_attention_mask: Optional[torch.FloatTensor] = None,
        ):
            for resnet, attn in zip(self.resnets, self.attentions):
                # pop res hidden states
                #print(f"in free crossatten upblock2d, hidden states shape: {hidden_states.shape}")
                res_hidden_states = res_hidden_states_tuple[-1]
                res_hidden_states_tuple = res_hidden_states_tuple[:-1]

                # --------------- FreeU code -----------------------
                # Only operate on the first two stages
                if hidden_states.shape[1] == 1280:
                    hidden_mean = hidden_states.mean(1).unsqueeze(1)
                    B = hidden_mean.shape[0]
                    hidden_max, _ = torch.max(hidden_mean.view(B, -1), dim=-1, keepdim=True) 
                    hidden_min, _ = torch.min(hidden_mean.view(B, -1), dim=-1, keepdim=True)

                    hidden_mean = (hidden_mean - hidden_min.unsqueeze(2).unsqueeze(3)) / (hidden_max - hidden_min).unsqueeze(2).unsqueeze(3)
                    
                    hidden_states[:,:640] = hidden_states[:,:640] * ((self.b1 - 1 ) * hidden_mean + 1)
                    res_hidden_states = Fourier_filter(res_hidden_states, threshold=1, scale=self.s1)
                if hidden_states.shape[1] == 640:
                    hidden_mean = hidden_states.mean(1).unsqueeze(1)
                    B = hidden_mean.shape[0]
                    hidden_max, _ = torch.max(hidden_mean.view(B, -1), dim=-1, keepdim=True) 
                    hidden_min, _ = torch.min(hidden_mean.view(B, -1), dim=-1, keepdim=True)
                    hidden_mean = (hidden_mean - hidden_min.unsqueeze(2).unsqueeze(3)) / (hidden_max - hidden_min).unsqueeze(2).unsqueeze(3)
                    
                    hidden_states[:,:320] = hidden_states[:,:320] * ((self.b2 - 1 ) * hidden_mean + 1)
                    res_hidden_states = Fourier_filter(res_hidden_states, threshold=1, scale=self.s2)
                # ---------------------------------------------------------

                hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)

                if self.training and self.gradient_checkpointing:

                    def create_custom_forward(module, return_dict=None):
                        def custom_forward(*inputs):
                            if return_dict is not None:
                                return module(*inputs, return_dict=return_dict)
                            else:
                                return module(*inputs)

                        return custom_forward

                    ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
                    hidden_states = torch.utils.checkpoint.checkpoint(
                        create_custom_forward(resnet),
                        hidden_states,
                        temb,
                        **ckpt_kwargs,
                    )
                    hidden_states = torch.utils.checkpoint.checkpoint(
                        create_custom_forward(attn, return_dict=False),
                        hidden_states,
                        encoder_hidden_states,
                        None,  # timestep
                        None,  # class_labels
                        cross_attention_kwargs,
                        attention_mask,
                        encoder_attention_mask,
                        **ckpt_kwargs,
                    )[0]
                else:
                    hidden_states = resnet(hidden_states, temb)
                    # hidden_states = attn(
                    #     hidden_states,
                    #     encoder_hidden_states=encoder_hidden_states,
                    #     cross_attention_kwargs=cross_attention_kwargs,
                    #     encoder_attention_mask=encoder_attention_mask,
                    #     return_dict=False,
                    # )[0]
                    hidden_states = attn(
                        hidden_states,
                        encoder_hidden_states=encoder_hidden_states,
                        cross_attention_kwargs=cross_attention_kwargs,
                    )[0]

            if self.upsamplers is not None:
                for upsampler in self.upsamplers:
                    hidden_states = upsampler(hidden_states, upsample_size)

            return hidden_states
        
        return forward
    
    for i, upsample_block in enumerate(model.unet.up_blocks):
        if isinstance_str(upsample_block, "CrossAttnUpBlock2D"):
            upsample_block.forward = up_forward(upsample_block)
            setattr(upsample_block, 'b1', b1)
            setattr(upsample_block, 'b2', b2)
            setattr(upsample_block, 's1', s1)
            setattr(upsample_block, 's2', s2)