File size: 10,956 Bytes
bd9da36
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""ViTDet backbone adapted from Detectron2"""

from functools import partial
from typing import List, Tuple, Union

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

from sam2.modeling.backbones.utils import (
    PatchEmbed,
    window_partition,
    window_unpartition,
    get_abs_pos,
)

from sam2.modeling.sam2_utils import DropPath, MLP, LayerScale

from functools import partial


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.
            attn_type: Type of attention operation, e.g. "vanilla", "vanilla-xformer".
        """
        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

    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)

        q = q.view(B, self.num_heads, H * W, -1)
        k = k.view(B, self.num_heads, H * W, -1)
        v = v.view(B, self.num_heads, H * W, -1)
        with torch.backends.cuda.sdp_kernel(
            enable_flash=True,
            enable_math=True,
            enable_mem_efficient=True,
        ):
            x = F.scaled_dot_product_attention(q, k, v)
        x = (
            x.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 Block(nn.Module):
    """Transformer blocks with support of window attention"""

    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,
        input_size=None,
        dropout=0.0,
        init_values=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.
            input_size (int or None): Input resolution for calculating the relative positional
                parameter size.
            dropout (float): Dropout rate.
        """
        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),
        )
        self.ls1 = (
            LayerScale(dim, init_values=init_values) if init_values else nn.Identity()
        )
        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else nn.Identity()

        self.norm2 = norm_layer(dim)
        self.mlp = MLP(
            dim,
            int(dim * mlp_ratio),
            dim,
            num_layers=2,
            activation=act_layer,
        )
        # self.mlp = Mlp2(
        #     in_features=dim,
        #     hidden_features=int(dim * mlp_ratio),
        #     act_layer=act_layer,
        #     drop=(dropout, 0.0),
        # )
        self.ls2 = (
            LayerScale(dim, init_values=init_values) if init_values else nn.Identity()
        )
        self.dropout = nn.Dropout(dropout)
        self.window_size = window_size

    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.ls1(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.dropout(self.drop_path(x))
        x = x + self.dropout(self.drop_path(self.ls2(self.mlp(self.norm2(x)))))

        return x


class ViT(nn.Module):
    """
    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=partial(nn.LayerNorm, eps=1e-6),
        act_layer=nn.GELU,
        use_abs_pos=True,
        use_rel_pos=False,
        rel_pos_zero_init=True,
        window_size=14,
        window_block_indexes=(0, 1, 3, 4, 6, 7, 9, 10),
        use_act_checkpoint=False,
        pretrain_img_size=224,
        pretrain_use_cls_token=True,
        dropout=0.0,
        weights_path=None,
        return_interm_layers=False,
        init_values=None,
    ):
        """
        Args:
            img_size (int): Input image size. Only relevant for rel pos.
            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.
            dropout (float): Dropout rate. Applied in residual blocks of attn, mlp and inside the mlp.
            path (str or None): Path to the pretrained weights.
            return_interm_layers (bool): Whether to return intermediate layers (all global attention blocks).
            freezing (BackboneFreezingType): Type of freezing.
        """
        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),
            padding=(0, 0),
            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()
        self.full_attn_ids = []
        cur_stage = 1
        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,
                input_size=(img_size // patch_size, img_size // patch_size),
                dropout=dropout,
                init_values=init_values,
            )
            if i not in window_block_indexes:
                self.full_attn_ids.append(i)
                cur_stage += 1

            self.blocks.append(block)

        self.return_interm_layers = return_interm_layers
        self.channel_list = (
            [embed_dim] * len(self.full_attn_ids)
            if return_interm_layers
            else [embed_dim]
        )
    

    def forward(self, x: torch.Tensor) -> List[torch.Tensor]:

        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])
            )

        outputs = []
        for i, blk in enumerate(self.blocks):
            x = blk(x)
            if (i == self.full_attn_ids[-1]) or (
                self.return_interm_layers and i in self.full_attn_ids
            ):
                feats = x.permute(0, 3, 1, 2)
                outputs.append(feats)

        return outputs