File size: 16,498 Bytes
2aac0e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import logging
import math
import fvcore.nn.weight_init as weight_init
import torch
import torch.nn as nn

from detectron2.layers import CNNBlockBase, Conv2d, get_norm
from detectron2.modeling.backbone.fpn import _assert_strides_are_log2_contiguous
import torch.nn.functional as F

from detectron2.modeling import BACKBONE_REGISTRY, Backbone, ShapeSpec
from .utils import (
    PatchEmbed,
    add_decomposed_rel_pos,
    get_abs_pos,
    window_partition,
    window_unpartition,
)
from functools import partial
import torch.utils.checkpoint as checkpoint

logger = logging.getLogger(__name__)


__all__ = ["ViT", "SimpleFeaturePyramid", "get_vit_lr_decay_rate"]


class Attention(nn.Module):
    """Multi-head Attention block with relative position embeddings."""

    def __init__(
        self,
        dim,
        num_heads=8,
        qkv_bias=True,
        use_rel_pos=False,
        rel_pos_zero_init=True,
        input_size=None,
    ):
        """
        Args:
            dim (int): Number of input channels.
            num_heads (int): Number of attention heads.
            qkv_bias (bool:  If True, add a learnable bias to query, key, value.
            rel_pos (bool): If True, add relative positional embeddings to the attention map.
            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
            input_size (int or None): Input resolution for calculating the relative positional
                parameter size.
        """
        super().__init__()
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = head_dim**-0.5

        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.proj = nn.Linear(dim, dim)

        self.use_rel_pos = use_rel_pos
        if self.use_rel_pos:
            # initialize relative positional embeddings
            self.rel_pos_h = nn.Parameter(torch.zeros(2 * input_size[0] - 1, head_dim))
            self.rel_pos_w = nn.Parameter(torch.zeros(2 * input_size[1] - 1, head_dim))

            if not rel_pos_zero_init:
                nn.init.trunc_normal_(self.rel_pos_h, std=0.02)
                nn.init.trunc_normal_(self.rel_pos_w, std=0.02)

    def forward(self, x):
        B, H, W, _ = x.shape
        # qkv with shape (3, B, nHead, H * W, C)
        qkv = self.qkv(x).reshape(B, H * W, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
        # q, k, v with shape (B * nHead, H * W, C)
        q, k, v = qkv.reshape(3, B * self.num_heads, H * W, -1).unbind(0)

        with torch.backends.cuda.sdp_kernel(
            enable_flash=True, enable_math=False, enable_mem_efficient=True
        ):
            x = F.scaled_dot_product_attention(q, k, v)
        attn = (q * self.scale) @ k.transpose(-2, -1)

        if self.use_rel_pos:
            attn = add_decomposed_rel_pos(attn, q, self.rel_pos_h, self.rel_pos_w, (H, W), (H, W))

        attn = attn.softmax(dim=-1)
        x = (attn @ v).view(B, self.num_heads, H, W, -1).permute(0, 2, 3, 1, 4).reshape(B, H, W, -1)
        x = self.proj(x)

        return x


class ResBottleneckBlock(CNNBlockBase):
    """
    The standard bottleneck residual block without the last activation layer.
    It contains 3 conv layers with kernels 1x1, 3x3, 1x1.
    """

    def __init__(
        self,
        in_channels,
        out_channels,
        bottleneck_channels,
        norm="LN",
        act_layer=nn.GELU,
    ):
        """
        Args:
            in_channels (int): Number of input channels.
            out_channels (int): Number of output channels.
            bottleneck_channels (int): number of output channels for the 3x3
                "bottleneck" conv layers.
            norm (str or callable): normalization for all conv layers.
                See :func:`layers.get_norm` for supported format.
            act_layer (callable): activation for all conv layers.
        """
        super().__init__(in_channels, out_channels, 1)

        self.conv1 = Conv2d(in_channels, bottleneck_channels, 1, bias=False)
        self.norm1 = get_norm(norm, bottleneck_channels)
        self.act1 = act_layer()

        self.conv2 = Conv2d(
            bottleneck_channels,
            bottleneck_channels,
            3,
            padding=1,
            bias=False,
        )
        self.norm2 = get_norm(norm, bottleneck_channels)
        self.act2 = act_layer()

        self.conv3 = Conv2d(bottleneck_channels, out_channels, 1, bias=False)
        self.norm3 = get_norm(norm, out_channels)

        for layer in [self.conv1, self.conv2, self.conv3]:
            weight_init.c2_msra_fill(layer)
        for layer in [self.norm1, self.norm2]:
            layer.weight.data.fill_(1.0)
            layer.bias.data.zero_()
        # zero init last norm layer.
        self.norm3.weight.data.zero_()
        self.norm3.bias.data.zero_()

    def forward(self, x):
        out = x
        for layer in self.children():
            out = layer(out)

        out = x + out
        return out


class Block(nn.Module):
    """Transformer blocks with support of window attention and residual propagation blocks"""

    def __init__(
        self,
        dim,
        num_heads,
        mlp_ratio=4.0,
        qkv_bias=True,
        drop_path=0.0,
        norm_layer=nn.LayerNorm,
        act_layer=nn.GELU,
        use_rel_pos=False,
        rel_pos_zero_init=True,
        window_size=0,
        use_residual_block=False,
        input_size=None,
    ):
        """
        Args:
            dim (int): Number of input channels.
            num_heads (int): Number of attention heads in each ViT block.
            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
            qkv_bias (bool): If True, add a learnable bias to query, key, value.
            drop_path (float): Stochastic depth rate.
            norm_layer (nn.Module): Normalization layer.
            act_layer (nn.Module): Activation layer.
            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
            window_size (int): Window size for window attention blocks. If it equals 0, then not
                use window attention.
            use_residual_block (bool): If True, use a residual block after the MLP block.
            input_size (int or None): Input resolution for calculating the relative positional
                parameter size.
        """
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention(
            dim,
            num_heads=num_heads,
            qkv_bias=qkv_bias,
            use_rel_pos=use_rel_pos,
            rel_pos_zero_init=rel_pos_zero_init,
            input_size=input_size if window_size == 0 else (window_size, window_size),
        )

        from timm.models.layers import DropPath, Mlp

        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else nn.Identity()
        self.norm2 = norm_layer(dim)
        self.mlp = Mlp(in_features=dim, hidden_features=int(dim * mlp_ratio), act_layer=act_layer)

        self.window_size = window_size

        self.use_residual_block = use_residual_block
        if use_residual_block:
            # Use a residual block with bottleneck channel as dim // 2
            self.residual = ResBottleneckBlock(
                in_channels=dim,
                out_channels=dim,
                bottleneck_channels=dim // 2,
                norm="LN",
                act_layer=act_layer,
            )

    def forward(self, x):
        shortcut = x
        x = self.norm1(x)
        # Window partition
        if self.window_size > 0:
            H, W = x.shape[1], x.shape[2]
            x, pad_hw = window_partition(x, self.window_size)
        x = self.attn(x)
        # Reverse window partition
        if self.window_size > 0:
            x = window_unpartition(x, self.window_size, pad_hw, (H, W))

        x = shortcut + self.drop_path(x)
        x = x + self.drop_path(self.mlp(self.norm2(x)))

        if self.use_residual_block:
            x = self.residual(x.permute(0, 3, 1, 2)).permute(0, 2, 3, 1)

        return x


class ViT(Backbone):
    """
    This module implements Vision Transformer (ViT) backbone in :paper:`vitdet`.
    "Exploring Plain Vision Transformer Backbones for Object Detection",
    https://arxiv.org/abs/2203.16527
    """

    def __init__(
        self,
        img_size=1024,
        patch_size=16,
        in_chans=3,
        embed_dim=768,
        depth=12,
        num_heads=12,
        mlp_ratio=4.0,
        qkv_bias=True,
        drop_path_rate=0.0,
        norm_layer=nn.LayerNorm,
        act_layer=nn.GELU,
        use_abs_pos=True,
        use_rel_pos=False,
        rel_pos_zero_init=True,
        window_size=0,
        window_block_indexes=(),
        residual_block_indexes=(),
        use_act_checkpoint=False,
        pretrain_img_size=224,
        pretrain_use_cls_token=True,
        out_feature="last_feat",
    ):
        """
        Args:
            img_size (int): Input image size.
            patch_size (int): Patch size.
            in_chans (int): Number of input image channels.
            embed_dim (int): Patch embedding dimension.
            depth (int): Depth of ViT.
            num_heads (int): Number of attention heads in each ViT block.
            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
            qkv_bias (bool): If True, add a learnable bias to query, key, value.
            drop_path_rate (float): Stochastic depth rate.
            norm_layer (nn.Module): Normalization layer.
            act_layer (nn.Module): Activation layer.
            use_abs_pos (bool): If True, use absolute positional embeddings.
            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
            window_size (int): Window size for window attention blocks.
            window_block_indexes (list): Indexes for blocks using window attention.
            residual_block_indexes (list): Indexes for blocks using conv propagation.
            use_act_checkpoint (bool): If True, use activation checkpointing.
            pretrain_img_size (int): input image size for pretraining models.
            pretrain_use_cls_token (bool): If True, pretrainig models use class token.
            out_feature (str): name of the feature from the last block.
        """
        super().__init__()
        self.pretrain_use_cls_token = pretrain_use_cls_token

        self.patch_embed = PatchEmbed(
            kernel_size=(patch_size, patch_size),
            stride=(patch_size, patch_size),
            in_chans=in_chans,
            embed_dim=embed_dim,
        )

        if use_abs_pos:
            # Initialize absolute positional embedding with pretrain image size.
            num_patches = (pretrain_img_size // patch_size) * (pretrain_img_size // patch_size)
            num_positions = (num_patches + 1) if pretrain_use_cls_token else num_patches
            self.pos_embed = nn.Parameter(torch.zeros(1, num_positions, embed_dim))
        else:
            self.pos_embed = None

        # stochastic depth decay rule
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)]

        self.blocks = nn.ModuleList()
        for i in range(depth):
            block = Block(
                dim=embed_dim,
                num_heads=num_heads,
                mlp_ratio=mlp_ratio,
                qkv_bias=qkv_bias,
                drop_path=dpr[i],
                norm_layer=norm_layer,
                act_layer=act_layer,
                use_rel_pos=use_rel_pos,
                rel_pos_zero_init=rel_pos_zero_init,
                window_size=window_size if i in window_block_indexes else 0,
                use_residual_block=i in residual_block_indexes,
                input_size=(img_size // patch_size, img_size // patch_size),
            )
            if use_act_checkpoint:
                # TODO: use torch.utils.checkpoint
                from fairscale.nn.checkpoint import checkpoint_wrapper

                block = checkpoint_wrapper(block)
            self.blocks.append(block)

        self._out_feature_channels = {out_feature: embed_dim}
        self._out_feature_strides = {out_feature: patch_size}
        self._out_features = [out_feature]

        if self.pos_embed is not None:
            nn.init.trunc_normal_(self.pos_embed, std=0.02)

        # In our method, we don't use backbone feature with stride 4
        self.fpn1 = nn.Sequential(
            nn.ConvTranspose2d(embed_dim, embed_dim // 2, kernel_size=2, stride=2),
        )
        self.fpn2 = nn.Identity() 
        self.fpn3 = nn.MaxPool2d(kernel_size=2, stride=2)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            nn.init.trunc_normal_(m.weight, std=0.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)

    def forward(self, x):
        x = self.patch_embed(x)
        if self.pos_embed is not None:
            x = x + get_abs_pos(
                self.pos_embed, self.pretrain_use_cls_token, (x.shape[1], x.shape[2])
            )

        for blk in self.blocks:
            x = blk(x)
        xp = x.permute(0, 3, 1, 2) # (b, h, w, c) --> (b, c, h, w)
        
        features = []
        ops = [self.fpn1, self.fpn2, self.fpn3]
        for i in range(len(ops)):
            features.append(ops[i](xp))
        rets = {"res{}".format(u + 3): v for (u,v) in enumerate(features)}

        return rets



@BACKBONE_REGISTRY.register()
class D2ViT(ViT, Backbone):
    def __init__(self, cfg, input_shape):
        use_checkpoint = cfg.MODEL.VIT.USE_CHECKPOINT
        if cfg.MODEL.VIT.NAME == "ViT-Base":
            embed_dim=768
            depth=12
            drop_path_rate=0.1
            num_heads=12
        elif cfg.MODEL.VIT.NAME == "ViT-Large":
            embed_dim=1024
            depth=24
            drop_path_rate=0.4
            num_heads=16
        elif cfg.MODEL.VIT.NAME == "ViT-huge":
            embed_dim=1280
            depth=32
            drop_path_rate=0.5
            num_heads=16
        else:
            raise ValueError("Unsupported ViT name")
        super().__init__(
            img_size=1024,
            patch_size=16,
            in_chans=input_shape.channels,
            embed_dim=embed_dim,
            depth=depth,
            num_heads=num_heads,
            drop_path_rate=drop_path_rate,
            window_size=14,
            mlp_ratio=4,
            qkv_bias=True,
            norm_layer=partial(nn.LayerNorm, eps=1e-6),
            window_block_indexes=[
                # 2, 5, 8 11 for global attention
                0,
                1,
                3,
                4,
                6,
                7,
                9,
                10,
            ],
            residual_block_indexes=[],
            use_rel_pos=True,
            out_feature="last_feat",
            use_act_checkpoint=use_checkpoint)

        self._out_features = cfg.MODEL.VIT.OUT_FEATURES

        self._out_feature_strides = {
            "res3": 8,
            "res4": 16,
            "res5": 32,
        }
        self._out_feature_channels = {
            "res3": embed_dim // 2,
            "res4": embed_dim,
            "res5": embed_dim,
        }

    def forward(self, x):
        """
        Args:
            x: Tensor of shape (N,C,H,W). H, W must be a multiple of ``self.size_divisibility``.
        Returns:
            dict[str->Tensor]: names and the corresponding features
        """
        assert (
            x.dim() == 4
        ), f"SwinTransformer takes an input of shape (N, C, H, W). Got {x.shape} instead!"
        outputs = {}
        y = super().forward(x)
        for k in y.keys():
            if k in self._out_features:
                outputs[k] = y[k]
        return outputs

    def output_shape(self):
        return {
            name: ShapeSpec(
                channels=self._out_feature_channels[name], stride=self._out_feature_strides[name]
            )
            for name in self._out_features
        }

    @property
    def size_divisibility(self):
        return 32