Swin Transformer解析
Howard Yin 2022-06-12 12:22:02 人工智能TransformerSwinTransformer神经网络注意力机制
# 最上层
先看看最上层：
# 类型和初始化
初始化的部分有很详细的参数注释，不多解释。
class SwinTransformer(nn.Module):
r""" Swin Transformer
A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows` -
https://arxiv.org/pdf/2103.14030
Args:
img_size (int | tuple(int)): Input image size. Default 224
patch_size (int | tuple(int)): Patch size. Default: 4
in_chans (int): Number of input image channels. Default: 3
num_classes (int): Number of classes for classification head. Default: 1000
embed_dim (int): Patch embedding dimension. Default: 96
depths (tuple(int)): Depth of each Swin Transformer layer.
num_heads (tuple(int)): Number of attention heads in different layers.
window_size (int): Window size. Default: 7
mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4
qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True
qk_scale (float): Override default qk scale of head_dim ** -0.5 if set. Default: None
drop_rate (float): Dropout rate. Default: 0
attn_drop_rate (float): Attention dropout rate. Default: 0
drop_path_rate (float): Stochastic depth rate. Default: 0.1
norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm.
ape (bool): If True, add absolute position embedding to the patch embedding. Default: False
patch_norm (bool): If True, add normalization after patch embedding. Default: True
use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False
"""
def __init__(self, img_size=224, patch_size=4, in_chans=3, num_classes=1000,
embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24],
window_size=7, mlp_ratio=4., qkv_bias=True, qk_scale=None,
drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1,
norm_layer=nn.LayerNorm, ape=False, patch_norm=True,
use_checkpoint=False, **kwargs):
super().__init__()
self.num_classes = num_classes
self.num_layers = len(depths)
self.embed_dim = embed_dim
self.ape = ape
self.patch_norm = patch_norm
self.num_features = int(embed_dim * 2 ** (self.num_layers - 1))
self.mlp_ratio = mlp_ratio
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
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
初始化了 Patch Embedding 这个PatchEmbed函数应该就是 Patch Embedding 算法，后面重点介绍
# split image into non-overlapping patches
self.patch_embed = PatchEmbed(
img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim,
norm_layer=norm_layer if self.patch_norm else None)
num_patches = self.patch_embed.num_patches
patches_resolution = self.patch_embed.patches_resolution
self.patches_resolution = patches_resolution
# absolute position embedding
if self.ape:
self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim))
trunc_normal_(self.absolute_pos_embed, std=.02)
1
2
3
4
5
6
7
8
9
10
11
12
2
3
4
5
6
7
8
9
10
11
12
用了Dropout
self.pos_drop = nn.Dropout(p=drop_rate)
# stochastic depth
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule
1
2
3
4
2
3
4
初始化模型主干，BasicLayer函数就是 Swin Transformer 主体模块，后面重点介绍
# build layers
self.layers = nn.ModuleList()
for i_layer in range(self.num_layers):
layer = BasicLayer(dim=int(embed_dim * 2 ** i_layer),
input_resolution=(patches_resolution[0] // (2 ** i_layer),
patches_resolution[1] // (2 ** i_layer)),
depth=depths[i_layer],
num_heads=num_heads[i_layer],
window_size=window_size,
mlp_ratio=self.mlp_ratio,
qkv_bias=qkv_bias, qk_scale=qk_scale,
drop=drop_rate, attn_drop=attn_drop_rate,
drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])],
norm_layer=norm_layer,
downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,
use_checkpoint=use_checkpoint)
self.layers.append(layer)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
接下来是LayerNorm和线性Head，不多介绍；最后还有个初始化权重的，也不必多说
self.norm = norm_layer(self.num_features)
self.avgpool = nn.AdaptiveAvgPool1d(1)
self.head = nn.Linear(self.num_features, num_classes) if num_classes > 0 else nn.Identity()
self.apply(self._init_weights)
def _init_weights(self, m):
if isinstance(m, nn.Linear):
trunc_normal_(m.weight, std=.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)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
2
3
4
5
6
7
8
9
10
11
12
13
14
# 推断过程
- Patch Embedding
- Dropout
- Swin Transformer 主干
- LayerNorm
- 平均值池化
- Head
def forward_features(self, x):
x = self.patch_embed(x)
if self.ape:
x = x + self.absolute_pos_embed
x = self.pos_drop(x)
for layer in self.layers:
x = layer(x)
x = self.norm(x) # B L C
x = self.avgpool(x.transpose(1, 2)) # B C 1
x = torch.flatten(x, 1)
return x
def forward(self, x):
x = self.forward_features(x)
x = self.head(x)
return x
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
# Patch Embedding PatchEmbed
class PatchEmbed(nn.Module):
r""" Image to Patch Embedding
Args:
img_size (int): Image size. Default: 224.
patch_size (int): Patch token size. Default: 4.
in_chans (int): Number of input image channels. Default: 3.
embed_dim (int): Number of linear projection output channels. Default: 96.
norm_layer (nn.Module, optional): Normalization layer. Default: None
"""
def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
super().__init__()
img_size = to_2tuple(img_size)
patch_size = to_2tuple(patch_size)
patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]]
self.img_size = img_size
self.patch_size = patch_size
self.patches_resolution = patches_resolution
self.num_patches = patches_resolution[0] * patches_resolution[1]
self.in_chans = in_chans
self.embed_dim = embed_dim
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
Patch Embedding实际上就是一个卷积，输入patch_sizexpatch_sizexin_chans的Patch，输出1x1xembed_dim，然后把stride设成和patch_size一样大，于是就是对不重叠的Patch进行操作。
self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
if norm_layer is not None:
self.norm = norm_layer(embed_dim)
else:
self.norm = None
1
2
3
4
5
2
3
4
5
计算过程也没啥好讲的，执行Patch Embedding的那个卷积之后结果显然是img_size/patch_sizeximg_size/patch_sizexembed_dim的矩阵，用flatten展平成img_size^2/patch_size^2xembed_dim矩阵即可。
def forward(self, x):
B, C, H, W = x.shape
# FIXME look at relaxing size constraints
assert H == self.img_size[0] and W == self.img_size[1], \
f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
x = self.proj(x).flatten(2).transpose(1, 2) # B Ph*Pw C
if self.norm is not None:
x = self.norm(x)
return x
1
2
3
4
5
6
7
8
9
2
3
4
5
6
7
8
9
# 主体模块 BasicLayer
class BasicLayer(nn.Module):
""" A basic Swin Transformer layer for one stage.
Args:
dim (int): Number of input channels.
input_resolution (tuple[int]): Input resolution.
depth (int): Number of blocks.
num_heads (int): Number of attention heads.
window_size (int): Local window size.
mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
drop (float, optional): Dropout rate. Default: 0.0
attn_drop (float, optional): Attention dropout rate. Default: 0.0
drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
"""
def __init__(self, dim, input_resolution, depth, num_heads, window_size,
mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0.,
drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False):
super().__init__()
self.dim = dim
self.input_resolution = input_resolution
self.depth = depth
self.use_checkpoint = use_checkpoint
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
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
又套了一个SwinTransformerBlock还是没有触及核心，后面重点解析这个SwinTransformerBlock。
# build blocks
self.blocks = nn.ModuleList([
SwinTransformerBlock(dim=dim, input_resolution=input_resolution,
num_heads=num_heads, window_size=window_size,
shift_size=0 if (i % 2 == 0) else window_size // 2,
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias, qk_scale=qk_scale,
drop=drop, attn_drop=attn_drop,
drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
norm_layer=norm_layer)
for i in range(depth)])
1
2
3
4
5
6
7
8
9
10
11
2
3
4
5
6
7
8
9
10
11
这个downsample是外面来的函数，从最上层的调用看，传进来的是一个PatchMerging函数，后面也得解析一下。
# patch merging layer
if downsample is not None:
self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer)
else:
self.downsample = None
1
2
3
4
5
2
3
4
5
推断过程就是先SwinTransformerBlock再downsample。
def forward(self, x):
for blk in self.blocks:
if self.use_checkpoint:
x = checkpoint.checkpoint(blk, x)
else:
x = blk(x)
if self.downsample is not None:
x = self.downsample(x)
return x
1
2
3
4
5
6
7
8
9
2
3
4
5
6
7
8
9
# Patch Merging PatchMerging
这PatchMerging是Swin Transformer的核心创新点之一，虽然很短但很重要。
# 初始化
class PatchMerging(nn.Module):
r""" Patch Merging Layer.
Args:
input_resolution (tuple[int]): Resolution of input feature.
dim (int): Number of input channels.
norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
"""
def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm):
super().__init__()
self.input_resolution = input_resolution
self.dim = dim
1
2
3
4
5
6
7
8
9
10
11
12
2
3
4
5
6
7
8
9
10
11
12
这是LayerNorm和线性映射层，后面的Patch Merging操作会将4个Patch叠在一起，所以这里的输入是4xdim；然后按照论文讲的，Patch Merging将4个Patch叠在一起后再用线性层将通道数减半，所以这里的输出是2xdim
self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
self.norm = norm_layer(4 * dim)
1
2
2
有个注意点是到这里Patch已经是一个一维向量了，而这里的nn.Linear只与输入的dim有关而与Patch数量无关，根据调用链，这个dim=int(embed_dim * 2 ** i_layer)，embed_dim是一维Patch Embedding的长度，输出的i_layer是第几层。看上图可以知道，特征图channel数每过一个Swin Transformer层就翻倍，所以这个dim实际上也只和Patch Embedding的长度有关，这意味着每个Patch都是经过相同的线性计算。由于图像大小固定，Patch数量也是固定的，这里实际上可以给每个Patch单用一个线性层参数，但作者没有这么用。
# 推断过程
def forward(self, x):
"""
x: B, H*W, C
"""
H, W = self.input_resolution
B, L, C = x.shape
assert L == H * W, "input feature has wrong size"
assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even."
x = x.view(B, H, W, C)
1
2
3
4
5
6
7
8
9
10
2
3
4
5
6
7
8
9
10
最核心的就是下面这个，这个操作就是把相邻的2x2共4个Patch叠在一起，然后用线性层将通道数减半。
x0 = x[:, 0::2, 0::2, :] # B H/2 W/2 C
x1 = x[:, 1::2, 0::2, :] # B H/2 W/2 C
x2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C
x3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C
x = torch.cat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C
x = x.view(B, -1, 4 * C) # B H/2*W/2 4*C
x = self.norm(x)
x = self.reduction(x)
return x
1
2
3
4
5
6
7
8
9
10
11
2
3
4
5
6
7
8
9
10
11
# 主体模块 SwinTransformerBlock
# Window Partition/Reverse
在开始讲主体模块前，有必要先讲Window Partition/Reverse机制，即函数window_partition和window_reverse。
window_partition函数是用于对张量划分窗口，指定窗口大小。将原本的张量从NxHxWxC, 划分成num_windows*Bxwindow_sizexwindow_sizexC，其中num_windows=H*W / (window_size*window_size)，即窗口的个数。而window_reverse函数则是对应的逆过程。这两个函数会在后面的Window Attention用到。
window_partition函数长这样：
def window_partition(x, window_size):
"""
Args:
x: (B, H, W, C)
window_size (int): window size
Returns:
windows: (num_windows*B, window_size, window_size, C)
"""
B, H, W, C = x.shape
x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
return windows
1
2
3
4
5
6
7
8
9
10
11
12
2
3
4
5
6
7
8
9
10
11
12
可以看到：
- 首先是
view将输入的mask中的长宽维扩出来一个window_size维度，即从BxHxWxC扩成BxH/window_sizexwindow_sizexW/window_sizexwindow_sizexC - 然后
permute在W/window_size和window_size维度进行转置，变成BxH/window_sizexW/window_sizexwindow_sizexwindow_sizexC view和permute都不改变原始矩阵中的值，而用contiguous是把转置后的矩阵拷贝一份- 最后的
view(-1, window_size, window_size, C)把BxH/window_sizexW/window_size这几个维度都抹了，只剩下一堆window_sizexwindow_sizexC
def window_reverse(windows, window_size, H, W):
B = int(windows.shape[0] / (H * W / window_size / window_size))
x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
return x
1
2
3
4
5
2
3
4
5
# 初始化
class SwinTransformerBlock(nn.Module):
r""" Swin Transformer Block.
Args:
dim (int): Number of input channels.
input_resolution (tuple[int]): Input resulotion.
num_heads (int): Number of attention heads.
window_size (int): Window size.
shift_size (int): Shift size for SW-MSA.
mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
drop (float, optional): Dropout rate. Default: 0.0
attn_drop (float, optional): Attention dropout rate. Default: 0.0
drop_path (float, optional): Stochastic depth rate. Default: 0.0
act_layer (nn.Module, optional): Activation layer. Default: nn.GELU
norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
"""
def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0,
mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0.,
act_layer=nn.GELU, norm_layer=nn.LayerNorm):
super().__init__()
self.dim = dim
self.input_resolution = input_resolution
self.num_heads = num_heads
self.window_size = window_size
self.shift_size = shift_size
self.mlp_ratio = mlp_ratio
if min(self.input_resolution) <= self.window_size:
# if window size is larger than input resolution, we don't partition windows
self.shift_size = 0
self.window_size = min(self.input_resolution)
assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
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
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
这个WindowAttention显然就是论文里讲的在窗口内的Attention，后面细讲。
self.norm1 = norm_layer(dim)
self.attn = WindowAttention(
dim, window_size=to_2tuple(self.window_size), num_heads=num_heads,
qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
1
2
3
4
2
3
4
下面这是个带Dropout的全连接层，不多讲。
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
self.norm2 = norm_layer(dim)
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
1
2
3
4
2
3
4
# Shift Windows核心代码
所谓Shift Windows核心就是下面这段代码。
if self.shift_size > 0:
1
# Attention mask for SW-MSA
首先是“calculate attention mask for SW-MSA”？
# calculate attention mask for SW-MSA
H, W = self.input_resolution
img_mask = torch.zeros((1, H, W, 1)) # 1 H W 1
h_slices = (slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None))
w_slices = (slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None))
cnt = 0
for h in h_slices:
for w in w_slices:
img_mask[:, h, w, :] = cnt
cnt += 1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
2
3
4
5
6
7
8
9
10
11
12
13
14
15
从两个for循环看，这个img_mask应该是一个标记，标记了每个待滑动区域的编号。实际上等价于：
img_mask[:, 0:-window_size, 0:-window_size, :] = 0
img_mask[:, 0:-window_size, -window_size: -shift_size, :] = 1
img_mask[:, 0:-window_size, -shift_size : , :] = 2
img_mask[:, -window_size: -shift_size, 0:-window_size, :] = 3
img_mask[:, -window_size: -shift_size, -window_size: -shift_size, :] = 4
img_mask[:, -window_size: -shift_size, -shift_size : , :] = 5
img_mask[:, -shift_size : , 0:-window_size, :] = 6
img_mask[:, -shift_size : , -window_size: -shift_size, :] = 7
img_mask[:, -shift_size : , -shift_size : , :] = 8
1
2
3
4
5
6
7
8
9
2
3
4
5
6
7
8
9
标记大概是这样：
根据调用链进行查找，可以发现window_size是用户指定的，而input_resolution和shift_size都是计算得到：
input_resolution=(patches_resolution[0] // (2 ** i_layer), patches_resolution[1] // (2 ** i_layer)),
1
其中i_layer是BasicLayer的编号，说明这个input_resolution逐层减半，和论文说的一样。
shift_size=0 if (i % 2 == 0) else window_size // 2
1
其中i是SwinTransformerBlock在BasicLayer中的编号，说明这个shift操作是隔层进行的，对应论文中的这个图：
即每两个SwinTransformerBlock组成一个基本单元，中间有一个Shift操作。
这也说明了W-MSA实际上只是SW-MSA少了一个Shift操作。
接下来是一个window_partition操作：
mask_windows = window_partition(img_mask, self.window_size) # nW, window_size, window_size, 1
mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
1
2
2
然后给填了一下数据？-100和0？
attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))
else:
attn_mask = None
self.register_buffer("attn_mask", attn_mask)
1
2
3
4
5
6
2
3
4
5
6
可以看到是对上面那个mask进行的操作。但是这到底是什么意思呢？
# 推断过程
def forward(self, x):
H, W = self.input_resolution
B, L, C = x.shape
assert L == H * W, "input feature has wrong size"
shortcut = x
x = self.norm1(x)
x = x.view(B, H, W, C)
1
2
3
4
5
6
7
8
2
3
4
5
6
7
8
这里一个shift操作，直接用torch.roll完成。
# cyclic shift
if self.shift_size > 0:
shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
else:
shifted_x = x
1
2
3
4
5
2
3
4
5
torch.roll的原理是这样：
那显然这里对应论文里的这个Cyclic Shift操作：
但这个和前面的img_mask又有什么关系？
接下来是window_partition把输入特征图切成windows，不用多讲。
# partition windows
x_windows = window_partition(shifted_x, self.window_size) # nW*B, window_size, window_size, C
x_windows = x_windows.view(-1, self.window_size * self.window_size, C) # nW*B, window_size*window_size, C
1
2
3
2
3
再然后就是应用WindowAttention：
# W-MSA/SW-MSA
attn_windows = self.attn(x_windows, mask=self.attn_mask) # nW*B, window_size*window_size, C
1
2
2
再然后就是把切成windows的特征图还原回来：
# merge windows
attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
shifted_x = window_reverse(attn_windows, self.window_size, H, W) # B H' W' C
1
2
3
2
3
再然后就是把移位后的特征图还原回来：
# reverse cyclic shift
if self.shift_size > 0:
x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
else:
x = shifted_x
x = x.view(B, H * W, C)
1
2
3
4
5
6
2
3
4
5
6
最后处理一下Dropout：
x = shortcut + self.drop_path(x)
# FFN
x = x + self.drop_path(self.mlp(self.norm2(x)))
return x
1
2
3
4
5
6
2
3
4
5
6
# 总结
总结一下，这里显然是Swin Transformer中各种Shift Windows的核心代码，但是看完了还是没能理解这个Shift Windows到底是在干什么。且看看后面的代码在回头总结吧。
# 核心模块WindowAttention
# 初始化
class WindowAttention(nn.Module):
r""" Window based multi-head self attention (W-MSA) module with relative position bias.
It supports both of shifted and non-shifted window.
Args:
dim (int): Number of input channels.
window_size (tuple[int]): The height and width of the window.
num_heads (int): Number of attention heads.
qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set
attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0
proj_drop (float, optional): Dropout ratio of output. Default: 0.0
"""
def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.):
super().__init__()
self.dim = dim
self.window_size = window_size # Wh, Ww
self.num_heads = num_heads
head_dim = dim // num_heads
self.scale = qk_scale or head_dim ** -0.5
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
一大段的相对位置编码：
# define a parameter table of relative position bias
self.relative_position_bias_table = nn.Parameter(
torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)) # 2*Wh-1 * 2*Ww-1, nH
# get pair-wise relative position index for each token inside the window
coords_h = torch.arange(self.window_size[0])
coords_w = torch.arange(self.window_size[1])
coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww
coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww
relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww
relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2
relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0
relative_coords[:, :, 1] += self.window_size[1] - 1
relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww
self.register_buffer("relative_position_index", relative_position_index)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
几个线性层和Dropout啥的，不必多讲
self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
self.attn_drop = nn.Dropout(attn_drop)
self.proj = nn.Linear(dim, dim)
self.proj_drop = nn.Dropout(proj_drop)
trunc_normal_(self.relative_position_bias_table, std=.02)
self.softmax = nn.Softmax(dim=-1)
1
2
3
4
5
6
7
2
3
4
5
6
7
# 推断过程
def forward(self, x, mask=None):
"""
Args:
x: input features with shape of (num_windows*B, N, C)
mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
"""
B_, N, C = x.shape
qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple)
q = q * self.scale
attn = (q @ k.transpose(-2, -1))
1
2
3
4
5
6
7
8
9
10
11
12
2
3
4
5
6
7
8
9
10
11
12
这里加上了相对位置编码：
relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH
relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww, Wh*Ww
attn = attn + relative_position_bias.unsqueeze(0)
1
2
3
4
2
3
4
然后加上了mask？按照前面讲的，这个mask里面是-100和0，所以这里是什么意思？
将mask加到attention的计算结果，并进行softmax。mask的值设置为-100，softmax后就会忽略掉对应的值
if mask is not None:
nW = mask.shape[0]
attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
attn = attn.view(-1, self.num_heads, N, N)
attn = self.softmax(attn)
else:
attn = self.softmax(attn)
1
2
3
4
5
6
7
2
3
4
5
6
7
处理Dropout：
attn = self.attn_drop(attn)
1
处理最后一个线性层和Dropout：
x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
x = self.proj(x)
x = self.proj_drop(x)
return x
1
2
3
4
2
3
4
# 总结
前面那个mask居然是在这里用的？还是没懂。
# 总结
还是看看大佬总结的吧《图解Swin Transformer》