File size: 27,741 Bytes
5bec700
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
from functools import partial
from abc import abstractmethod
import torch
import torch.nn as nn
from einops import rearrange
import torch.nn.functional as F
from lvdm.models.utils_diffusion import timestep_embedding
from lvdm.common import checkpoint
from lvdm.basics import zero_module, conv_nd, linear, avg_pool_nd, normalization
from lvdm.modules.attention import SpatialTransformer, TemporalTransformer


class TimestepBlock(nn.Module):
    """
    Any module where forward() takes timestep embeddings as a second argument.
    """

    @abstractmethod
    def forward(self, x, emb):
        """
        Apply the module to `x` given `emb` timestep embeddings.
        """


class TimestepEmbedSequential(nn.Sequential, TimestepBlock):
    """
    A sequential module that passes timestep embeddings to the children that
    support it as an extra input.
    """

    def forward(self, x, emb, context=None, batch_size=None):
        for layer in self:
            if isinstance(layer, TimestepBlock):
                x = layer(x, emb, batch_size)
            elif isinstance(layer, SpatialTransformer):
                x = layer(x, context)
            elif isinstance(layer, TemporalTransformer):
                x = rearrange(x, "(b f) c h w -> b c f h w", b=batch_size)
                x = layer(x, context)
                x = rearrange(x, "b c f h w -> (b f) c h w")
            else:
                x = layer(
                    x,
                )
        return x


class Downsample(nn.Module):
    """
    A downsampling layer with an optional convolution.
    :param channels: channels in the inputs and outputs.
    :param use_conv: a bool determining if a convolution is applied.
    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
                 downsampling occurs in the inner-two dimensions.
    """

    def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1):
        super().__init__()
        self.channels = channels
        self.out_channels = out_channels or channels
        self.use_conv = use_conv
        self.dims = dims
        stride = 2 if dims != 3 else (1, 2, 2)
        if use_conv:
            self.op = conv_nd(
                dims,
                self.channels,
                self.out_channels,
                3,
                stride=stride,
                padding=padding,
            )
        else:
            assert self.channels == self.out_channels
            self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)

    def forward(self, x):
        assert x.shape[1] == self.channels
        return self.op(x)


class Upsample(nn.Module):
    """
    An upsampling layer with an optional convolution.
    :param channels: channels in the inputs and outputs.
    :param use_conv: a bool determining if a convolution is applied.
    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
                 upsampling occurs in the inner-two dimensions.
    """

    def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1):
        super().__init__()
        self.channels = channels
        self.out_channels = out_channels or channels
        self.use_conv = use_conv
        self.dims = dims
        if use_conv:
            self.conv = conv_nd(
                dims, self.channels, self.out_channels, 3, padding=padding
            )

    def forward(self, x):
        assert x.shape[1] == self.channels
        if self.dims == 3:
            x = F.interpolate(
                x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest"
            )
        else:
            x = F.interpolate(x, scale_factor=2, mode="nearest")
        if self.use_conv:
            x = self.conv(x)
        return x


class ResBlock(TimestepBlock):
    """
    A residual block that can optionally change the number of channels.
    :param channels: the number of input channels.
    :param emb_channels: the number of timestep embedding channels.
    :param dropout: the rate of dropout.
    :param out_channels: if specified, the number of out channels.
    :param use_conv: if True and out_channels is specified, use a spatial
        convolution instead of a smaller 1x1 convolution to change the
        channels in the skip connection.
    :param dims: determines if the signal is 1D, 2D, or 3D.
    :param up: if True, use this block for upsampling.
    :param down: if True, use this block for downsampling.
    """

    def __init__(
        self,
        channels,
        emb_channels,
        dropout,
        out_channels=None,
        use_scale_shift_norm=False,
        dims=2,
        use_checkpoint=False,
        use_conv=False,
        up=False,
        down=False,
        use_temporal_conv=False,
        tempspatial_aware=False,
    ):
        super().__init__()
        self.channels = channels
        self.emb_channels = emb_channels
        self.dropout = dropout
        self.out_channels = out_channels or channels
        self.use_conv = use_conv
        self.use_checkpoint = use_checkpoint
        self.use_scale_shift_norm = use_scale_shift_norm
        self.use_temporal_conv = use_temporal_conv

        self.in_layers = nn.Sequential(
            normalization(channels),
            nn.SiLU(),
            conv_nd(dims, channels, self.out_channels, 3, padding=1),
        )

        self.updown = up or down

        if up:
            self.h_upd = Upsample(channels, False, dims)
            self.x_upd = Upsample(channels, False, dims)
        elif down:
            self.h_upd = Downsample(channels, False, dims)
            self.x_upd = Downsample(channels, False, dims)
        else:
            self.h_upd = self.x_upd = nn.Identity()

        self.emb_layers = nn.Sequential(
            nn.SiLU(),
            nn.Linear(
                emb_channels,
                2 * self.out_channels if use_scale_shift_norm else self.out_channels,
            ),
        )
        self.out_layers = nn.Sequential(
            normalization(self.out_channels),
            nn.SiLU(),
            nn.Dropout(p=dropout),
            zero_module(nn.Conv2d(self.out_channels, self.out_channels, 3, padding=1)),
        )

        if self.out_channels == channels:
            self.skip_connection = nn.Identity()
        elif use_conv:
            self.skip_connection = conv_nd(
                dims, channels, self.out_channels, 3, padding=1
            )
        else:
            self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)

        if self.use_temporal_conv:
            self.temopral_conv = TemporalConvBlock(
                self.out_channels,
                self.out_channels,
                dropout=0.1,
                spatial_aware=tempspatial_aware,
            )

    def forward(self, x, emb, batch_size=None):
        """
        Apply the block to a Tensor, conditioned on a timestep embedding.
        :param x: an [N x C x ...] Tensor of features.
        :param emb: an [N x emb_channels] Tensor of timestep embeddings.
        :return: an [N x C x ...] Tensor of outputs.
        """
        input_tuple = (
            x,
            emb,
        )
        if batch_size:
            forward_batchsize = partial(self._forward, batch_size=batch_size)
            return checkpoint(
                forward_batchsize, input_tuple, self.parameters(), self.use_checkpoint
            )
        return checkpoint(
            self._forward, input_tuple, self.parameters(), self.use_checkpoint
        )

    def _forward(
        self,
        x,
        emb,
        batch_size=None,
    ):
        if self.updown:
            in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
            h = in_rest(x)
            h = self.h_upd(h)
            x = self.x_upd(x)
            h = in_conv(h)
        else:
            h = self.in_layers(x)
        emb_out = self.emb_layers(emb).type(h.dtype)
        while len(emb_out.shape) < len(h.shape):
            emb_out = emb_out[..., None]
        if self.use_scale_shift_norm:
            out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
            scale, shift = torch.chunk(emb_out, 2, dim=1)
            h = out_norm(h) * (1 + scale) + shift
            h = out_rest(h)
        else:
            h = h + emb_out
            h = self.out_layers(h)
        h = self.skip_connection(x) + h

        if self.use_temporal_conv and batch_size:
            h = rearrange(h, "(b t) c h w -> b c t h w", b=batch_size)
            h = self.temopral_conv(h)
            h = rearrange(h, "b c t h w -> (b t) c h w")
        return h


class TemporalConvBlock(nn.Module):
    """
    Adapted from modelscope: https://github.com/modelscope/modelscope/blob/master/modelscope/models/multi_modal/video_synthesis/unet_sd.py
    """

    def __init__(
        self, in_channels, out_channels=None, dropout=0.0, spatial_aware=False
    ):
        super(TemporalConvBlock, self).__init__()
        if out_channels is None:
            out_channels = in_channels
        self.in_channels = in_channels
        self.out_channels = out_channels
        kernel_shape = (3, 1, 1) if not spatial_aware else (3, 3, 3)
        padding_shape = (1, 0, 0) if not spatial_aware else (1, 1, 1)

        # conv layers
        self.conv1 = nn.Sequential(
            nn.GroupNorm(32, in_channels),
            nn.SiLU(),
            nn.Conv3d(in_channels, out_channels, kernel_shape, padding=padding_shape),
        )
        self.conv2 = nn.Sequential(
            nn.GroupNorm(32, out_channels),
            nn.SiLU(),
            nn.Dropout(dropout),
            nn.Conv3d(out_channels, in_channels, kernel_shape, padding=padding_shape),
        )
        self.conv3 = nn.Sequential(
            nn.GroupNorm(32, out_channels),
            nn.SiLU(),
            nn.Dropout(dropout),
            nn.Conv3d(out_channels, in_channels, (3, 1, 1), padding=(1, 0, 0)),
        )
        self.conv4 = nn.Sequential(
            nn.GroupNorm(32, out_channels),
            nn.SiLU(),
            nn.Dropout(dropout),
            nn.Conv3d(out_channels, in_channels, (3, 1, 1), padding=(1, 0, 0)),
        )

        # zero out the last layer params,so the conv block is identity
        nn.init.zeros_(self.conv4[-1].weight)
        nn.init.zeros_(self.conv4[-1].bias)

    def forward(self, x):
        identity = x
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        x = self.conv4(x)

        return x + identity


class UNetModel(nn.Module):
    """
    The full UNet model with attention and timestep embedding.
    :param in_channels: in_channels in the input Tensor.
    :param model_channels: base channel count for the model.
    :param out_channels: channels in the output Tensor.
    :param num_res_blocks: number of residual blocks per downsample.
    :param attention_resolutions: a collection of downsample rates at which
        attention will take place. May be a set, list, or tuple.
        For example, if this contains 4, then at 4x downsampling, attention
        will be used.
    :param dropout: the dropout probability.
    :param channel_mult: channel multiplier for each level of the UNet.
    :param conv_resample: if True, use learned convolutions for upsampling and
        downsampling.
    :param dims: determines if the signal is 1D, 2D, or 3D.
    :param num_classes: if specified (as an int), then this model will be
        class-conditional with `num_classes` classes.
    :param use_checkpoint: use gradient checkpointing to reduce memory usage.
    :param num_heads: the number of attention heads in each attention layer.
    :param num_heads_channels: if specified, ignore num_heads and instead use
                               a fixed channel width per attention head.
    :param num_heads_upsample: works with num_heads to set a different number
                               of heads for upsampling. Deprecated.
    :param use_scale_shift_norm: use a FiLM-like conditioning mechanism.
    :param resblock_updown: use residual blocks for up/downsampling.
    """

    def __init__(
        self,
        in_channels,
        model_channels,
        out_channels,
        num_res_blocks,
        attention_resolutions,
        dropout=0.0,
        channel_mult=(1, 2, 4, 8),
        conv_resample=True,
        dims=2,
        context_dim=None,
        use_scale_shift_norm=False,
        resblock_updown=False,
        num_heads=-1,
        num_head_channels=-1,
        transformer_depth=1,
        use_linear=False,
        use_checkpoint=False,
        temporal_conv=False,
        tempspatial_aware=False,
        temporal_attention=True,
        temporal_selfatt_only=True,
        use_relative_position=True,
        use_causal_attention=False,
        temporal_length=None,
        use_fp16=False,
        addition_attention=False,
        use_image_attention=False,
        temporal_transformer_depth=1,
        fps_cond=False,
        time_cond_proj_dim=None,
        motion_cond_proj_dim=None,
        record_attn_probs=False,
    ):
        super(UNetModel, self).__init__()
        if num_heads == -1:
            assert (
                num_head_channels != -1
            ), "Either num_heads or num_head_channels has to be set"
        if num_head_channels == -1:
            assert (
                num_heads != -1
            ), "Either num_heads or num_head_channels has to be set"

        self.in_channels = in_channels
        self.model_channels = model_channels
        self.out_channels = out_channels
        self.num_res_blocks = num_res_blocks
        self.attention_resolutions = attention_resolutions
        self.dropout = dropout
        self.channel_mult = channel_mult
        self.conv_resample = conv_resample
        self.temporal_attention = temporal_attention
        time_embed_dim = model_channels * 4
        self.use_checkpoint = use_checkpoint
        self.dtype = torch.float16 if use_fp16 else torch.float32
        self.addition_attention = addition_attention
        self.use_image_attention = use_image_attention
        self.fps_cond = fps_cond
        self.time_cond_proj_dim = time_cond_proj_dim
        self.motion_cond_proj_dim = motion_cond_proj_dim

        self.time_embed = nn.Sequential(
            linear(model_channels, time_embed_dim),
            nn.SiLU(),
            linear(time_embed_dim, time_embed_dim),
        )
        if self.fps_cond:
            self.fps_embedding = nn.Sequential(
                linear(model_channels, time_embed_dim),
                nn.SiLU(),
                linear(time_embed_dim, time_embed_dim),
            )
        if time_cond_proj_dim is not None:
            self.time_cond_proj = nn.Linear(
                time_cond_proj_dim, model_channels, bias=False
            )
        else:
            self.time_cond_proj = None
        
        if motion_cond_proj_dim is not None:
            self.motion_cond_proj = nn.Linear(
                motion_cond_proj_dim, model_channels, bias=False
            )
            self.combine_proj = nn.Linear(
                model_channels * 2, model_channels, bias=False
            )
        else:
            self.motion_cond_proj = None
            self.combine_proj = None

        self.input_blocks = nn.ModuleList(
            [
                TimestepEmbedSequential(
                    conv_nd(dims, in_channels, model_channels, 3, padding=1)
                )
            ]
        )
        if self.addition_attention:
            self.init_attn = TimestepEmbedSequential(
                TemporalTransformer(
                    model_channels,
                    n_heads=8,
                    d_head=num_head_channels,
                    depth=transformer_depth,
                    context_dim=context_dim,
                    use_checkpoint=use_checkpoint,
                    only_self_att=temporal_selfatt_only,
                    causal_attention=use_causal_attention,
                    relative_position=use_relative_position,
                    temporal_length=temporal_length,
                )
            )

        input_block_chans = [model_channels]
        ch = model_channels
        ds = 1
        for level, mult in enumerate(channel_mult):
            for _ in range(num_res_blocks):
                layers = [
                    ResBlock(
                        ch,
                        time_embed_dim,
                        dropout,
                        out_channels=mult * model_channels,
                        dims=dims,
                        use_checkpoint=use_checkpoint,
                        use_scale_shift_norm=use_scale_shift_norm,
                        tempspatial_aware=tempspatial_aware,
                        use_temporal_conv=temporal_conv,
                    )
                ]
                ch = mult * model_channels
                if ds in attention_resolutions:
                    if num_head_channels == -1:
                        dim_head = ch // num_heads
                    else:
                        num_heads = ch // num_head_channels
                        dim_head = num_head_channels
                    layers.append(
                        SpatialTransformer(
                            ch,
                            num_heads,
                            dim_head,
                            depth=transformer_depth,
                            context_dim=context_dim,
                            use_linear=use_linear,
                            use_checkpoint=use_checkpoint,
                            disable_self_attn=False,
                            img_cross_attention=self.use_image_attention,
                        )
                    )
                    if self.temporal_attention:
                        layers.append(
                            TemporalTransformer(
                                ch,
                                num_heads,
                                dim_head,
                                depth=temporal_transformer_depth,
                                context_dim=context_dim,
                                use_linear=use_linear,
                                use_checkpoint=use_checkpoint,
                                only_self_att=temporal_selfatt_only,
                                causal_attention=use_causal_attention,
                                relative_position=use_relative_position,
                                temporal_length=temporal_length,
                            )
                        )
                self.input_blocks.append(TimestepEmbedSequential(*layers))
                input_block_chans.append(ch)
            if level != len(channel_mult) - 1:
                out_ch = ch
                self.input_blocks.append(
                    TimestepEmbedSequential(
                        ResBlock(
                            ch,
                            time_embed_dim,
                            dropout,
                            out_channels=out_ch,
                            dims=dims,
                            use_checkpoint=use_checkpoint,
                            use_scale_shift_norm=use_scale_shift_norm,
                            down=True,
                        )
                        if resblock_updown
                        else Downsample(
                            ch, conv_resample, dims=dims, out_channels=out_ch
                        )
                    )
                )
                ch = out_ch
                input_block_chans.append(ch)
                ds *= 2

        if num_head_channels == -1:
            dim_head = ch // num_heads
        else:
            num_heads = ch // num_head_channels
            dim_head = num_head_channels
        layers = [
            ResBlock(
                ch,
                time_embed_dim,
                dropout,
                dims=dims,
                use_checkpoint=use_checkpoint,
                use_scale_shift_norm=use_scale_shift_norm,
                tempspatial_aware=tempspatial_aware,
                use_temporal_conv=temporal_conv,
            ),
            SpatialTransformer(
                ch,
                num_heads,
                dim_head,
                depth=transformer_depth,
                context_dim=context_dim,
                use_linear=use_linear,
                use_checkpoint=use_checkpoint,
                disable_self_attn=False,
                img_cross_attention=self.use_image_attention,
            ),
        ]
        if self.temporal_attention:
            layers.append(
                TemporalTransformer(
                    ch,
                    num_heads,
                    dim_head,
                    depth=temporal_transformer_depth,
                    context_dim=context_dim,
                    use_linear=use_linear,
                    use_checkpoint=use_checkpoint,
                    only_self_att=temporal_selfatt_only,
                    causal_attention=use_causal_attention,
                    relative_position=use_relative_position,
                    temporal_length=temporal_length,
                )
            )
        layers.append(
            ResBlock(
                ch,
                time_embed_dim,
                dropout,
                dims=dims,
                use_checkpoint=use_checkpoint,
                use_scale_shift_norm=use_scale_shift_norm,
                tempspatial_aware=tempspatial_aware,
                use_temporal_conv=temporal_conv,
            )
        )
        self.middle_block = TimestepEmbedSequential(*layers)

        self.output_blocks = nn.ModuleList([])
        for level, mult in list(enumerate(channel_mult))[::-1]:
            for i in range(num_res_blocks + 1):
                ich = input_block_chans.pop()
                layers = [
                    ResBlock(
                        ch + ich,
                        time_embed_dim,
                        dropout,
                        out_channels=mult * model_channels,
                        dims=dims,
                        use_checkpoint=use_checkpoint,
                        use_scale_shift_norm=use_scale_shift_norm,
                        tempspatial_aware=tempspatial_aware,
                        use_temporal_conv=temporal_conv,
                    )
                ]
                ch = model_channels * mult
                if ds in attention_resolutions:
                    if num_head_channels == -1:
                        dim_head = ch // num_heads
                    else:
                        num_heads = ch // num_head_channels
                        dim_head = num_head_channels
                    layers.append(
                        SpatialTransformer(
                            ch,
                            num_heads,
                            dim_head,
                            depth=transformer_depth,
                            context_dim=context_dim,
                            use_linear=use_linear,
                            use_checkpoint=use_checkpoint,
                            disable_self_attn=False,
                            img_cross_attention=self.use_image_attention,
                        )
                    )
                    if self.temporal_attention:
                        layers.append(
                            TemporalTransformer(
                                ch,
                                num_heads,
                                dim_head,
                                depth=temporal_transformer_depth,
                                context_dim=context_dim,
                                use_linear=use_linear,
                                use_checkpoint=use_checkpoint,
                                only_self_att=temporal_selfatt_only,
                                causal_attention=use_causal_attention,
                                relative_position=use_relative_position,
                                temporal_length=temporal_length,
                                record_attn_probs=record_attn_probs,
                            )
                        )
                if level and i == num_res_blocks:
                    out_ch = ch
                    layers.append(
                        ResBlock(
                            ch,
                            time_embed_dim,
                            dropout,
                            out_channels=out_ch,
                            dims=dims,
                            use_checkpoint=use_checkpoint,
                            use_scale_shift_norm=use_scale_shift_norm,
                            up=True,
                        )
                        if resblock_updown
                        else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch)
                    )
                    ds //= 2
                self.output_blocks.append(TimestepEmbedSequential(*layers))

        self.out = nn.Sequential(
            normalization(ch),
            nn.SiLU(),
            zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)),
        )

    def forward(
        self,
        x,
        timesteps,
        context=None,
        features_adapter=None,
        fps=16,
        timestep_cond=None,
        motion_cond=None,
        **kwargs
    ):
        t_emb = timestep_embedding(
            timesteps, self.model_channels, repeat_only=False
        ).to(self.dtype)
        if timestep_cond is not None:
            timestep_cond_embed = self.time_cond_proj(timestep_cond)
        else:
            timestep_cond_embed = 0.
        if motion_cond is not None:
            assert timestep_cond is not None
            motion_cond_emb = self.motion_cond_proj(motion_cond)
            combined_cond_emb = self.combine_proj(
                torch.cat([timestep_cond_embed, motion_cond_emb], dim=1)
            )
        else:
            combined_cond_emb = timestep_cond_embed
        emb = self.time_embed(t_emb + combined_cond_emb)

        if self.fps_cond:
            if type(fps) == int:
                fps = torch.full_like(timesteps, fps)
            fps_emb = timestep_embedding(
                fps, self.model_channels, repeat_only=False
            ).to(self.dtype)
            emb += self.fps_embedding(fps_emb)

        b, _, t, _, _ = x.shape
        ## repeat t times for context [(b t) 77 768] & time embedding
        context = context.repeat_interleave(repeats=t, dim=0)
        emb = emb.repeat_interleave(repeats=t, dim=0)

        ## always in shape (b t) c h w, except for temporal layer
        x = rearrange(x, "b c t h w -> (b t) c h w")

        h = x.type(self.dtype)
        adapter_idx = 0
        hs = []
        for id, module in enumerate(self.input_blocks):
            h = module(h, emb, context=context, batch_size=b)
            if id == 0 and self.addition_attention:
                h = self.init_attn(h, emb, context=context, batch_size=b)
            ## plug-in adapter features
            if ((id + 1) % 3 == 0) and features_adapter is not None:
                h = h + features_adapter[adapter_idx]
                adapter_idx += 1
            hs.append(h)
        if features_adapter is not None:
            assert len(features_adapter) == adapter_idx, "Wrong features_adapter"

        h = self.middle_block(h, emb, context=context, batch_size=b)
        for module in self.output_blocks:
            h = torch.cat([h, hs.pop()], dim=1)
            h = module(h, emb, context=context, batch_size=b)
        h = h.type(x.dtype)
        y = self.out(h)

        # reshape back to (b c t h w)
        y = rearrange(y, "(b t) c h w -> b c t h w", b=b)
        return y