swin transformer论文及代码学习

Posted 2021-06-14 若水菱花

论文    源码

Overall Architecture

Image

输入image的大小为三维矩阵：H W 3。H为Height，W为Width，3为通道channel，这里指的是RGB。图中只以一个image为例，也就是batch_size = 1。

Patch Partition

一张图片读入后表示为像素矩阵，需要先对图片进行patch partition处理，将图片的最小单位从像素转变为patch。论文中所给的示例为一个patch由4*4个pixel构成，即patch partition模块用包含4*4个像素的patch来对像素矩阵进行分割，并一个patch中的像素值合成一个向量。输入的像素矩阵经过处理后变为的三维矩阵，其中H/4 * W/4表示patch的数量，48为channel，由3*4*4得来。

Linear Embedding

A linear embedding layer is applied on this raw-valued feature to project it to an arbitrary dimension(denoted as C).C default 96.

代码中PatchEmbed()类中包含了Patch Partition 和Linear Embedding两个模块。具体代码含义见下面注释。

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. 4 pixel
        in_chans (int): Number of input image channels. Default: 3. RGB
        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 -> tuple(img_size,img_size) 即（224，224）
        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
        #图片中的patch个数
        self.num_patches = patches_resolution[0] * patches_resolution[1]
        #input channel default RGB:3
        self.in_chans = in_chans
        #output channel
        self.embed_dim = embed_dim
        #in_chans:3 out_chans:96 即输入一个图片的三个通道 经过4*4的卷积核和步长为4后输出channel为96
        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
        #默认经过embedding后进行标准化norm layer
        if norm_layer is not None:
            self.norm = norm_layer(embed_dim)
        else:
            self.norm = None

    def forward(self, x):
        # batch channel height width
        B, C, H, W = x.shape
        # FIXME look at relaxing size constraints
        #判断输入图片格式是否符合模型
        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]})."
        #将图片进行embed flatten将数组从括号中的维度开始将后面的所有数降为一维 
        # 这里x为B C PH PW 4维
        #经过flatten变为B C PH*PW 3维 再经过transpose调换维度 变为B PH*PW C
        x = self.proj(x).flatten(2).transpose(1, 2)  # B Ph*Pw C
        #标准化
        if self.norm is not None:
            x = self.norm(x)
        return x
    #理论计算复杂度
    def flops(self):
        #用patch的数量来进行计算
        Ho, Wo = self.patches_resolution
        # 图片patch的数量 网络输入输出通道 每个patch内含有的pixel
        flops = Ho * Wo * (self.embed_dim * self.in_chans) * (self.patch_size[0] * self.patch_size[1])
        #如果未进行标准化 则再加上
        if self.norm is not None:
            flops += Ho * Wo * self.embed_dim
        return flops

Patch Merging

实现了patch合并,使得channel*4，在经过卷积网络进行降维,使得channel/2。

The first patch merging layer concatenates the features of each group of 2*2 neighboring patches,and applies a linear layer on the 4C-dimensional concatenated features.This reduces the number of tokens by a multiple of 2*2 = 4(2*downsampling of resolution),and the output dimension is set to 2C.

class PatchMerging(nn.Module):
    r""" Patch Merging Layer.

    Args:
        input_resolution (tuple[int]): Resolution of input feature. feature map的size
        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
        #input channel:4*dim output channel:2*dim 降维
        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
        self.norm = norm_layer(4 * dim)

    def forward(self, x):
        """
        x: B, H*W, C
        """
        H, W = self.input_resolution
        B, L, C = x.shape
        #检验输入是否符合模型
        assert L == H * W, "input feature has wrong size"
        #判断高 宽是否为偶数
        assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even."
        #reshape
        x = x.view(B, H, W, C)
        #0::2 表示从0开始step为2，一直取完
        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
        #cat为接合张量 -1表示最内侧的那一维 如0，1，2，3 则-1表示3 这里表示xi张量按照C这一维度进行接合
        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
        # 标准化
        x = self.norm(x)
        x = self.reduction(x)
        return x
    #一些额外的信息
    def extra_repr(self) -> str:
        return f"input_resolution={self.input_resolution}, dim={self.dim}"
    #计算复杂度
    def flops(self):
        H, W = self.input_resolution
        flops = H * W * self.dim
        flops += (H // 2) * (W // 2) * 4 * self.dim * 2 * self.dim
        return flops

Swin Transformer Block

LN  LayerNorm

A LayerNorm layer is applied before each MSA module and each MLP,and a residual connection is applied after each module.

MLP  Muti-Layer Perception

就是一个两层fc网络

class Mlp(nn.Module):
    """
    Muti－Layer Perception ，MLP
    就是一个两层fc网络
    in_features:
    hidden_features:
    out_features:
    act_layer: activation function
    """
    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
        """
        GELU(x)=x∗Φ(x)
        """
        super().__init__()
        #若未输入out_features hidden_features则输入输出维度一样
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        #两个fc网络
        self.fc1 = nn.Linear(in_features, hidden_features)
        self.act = act_layer()
        self.fc2 = nn.Linear(hidden_features, out_features)
        #随机赋0 防止过拟合
        self.drop = nn.Dropout(drop)

    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        #随机赋0
        x = self.drop(x)
        x = self.fc2(x)
        x = self.drop(x)
        return x

W-MSA  and  SW-MSA

W-MSA and SW-MSA are multi-head self attention modules with regular and shifed windowing configurations,respectively.

这里的multi-head self attention 请见其他文章。下面主要介绍window_partition，shifted window partition and window_attention

这一部分借鉴图解Swin Transformer zzzk 受益匪浅。

window_partition and window_reverse

用window分割multi-head self attention的输入

#将输入: batch height width c-dimension 分割为: num_windows*batch window_size window_size c-dimension
#其中num_windows=height*width/(window_size*window_size)
def window_partition(x, window_size):
    """
    Args:
        x: (B, H, W, C) batch height width c-dimention
        window_size (int): window size

    Returns:
        windows: (num_windows*B, window_size, window_size, C)
    """
    B, H, W, C = x.shape
    # a//b 整除
    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
    # permute 将3和2调换 contiguous检测在内存上存储是否连续
    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
    return windows

#是window_partition的逆过程
def window_reverse(windows, window_size, H, W):
    """
    Args:
        windows: (num_windows*B, window_size, window_size, C)
        window_size (int): Window size
        H (int): Height of image
        W (int): Width of image

    Returns:
        x: (B, H, W, C)
    """
    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

window_attention

以window为单位求输入通过multi-head self attention后的输出

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
        #输入的dim被multi-head attention的每个head瓜分 各自进行attention计算
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5

        # define a parameter table of relative position bias
        #初始化一个全零矩阵 (2*wh-1)*(2*ww-1) num_heads  相对位置在数轴上的分布维[-window_size + 1,window_size - 1]
        #里面的参数需要进行学习
        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
        #生成window中的每个token的相对位置，其中最小的相对位置为0，后面需要用这个相对位置作为索引，在relative_position_bias中去找到相应的bias
        coords_h = torch.arange(self.window_size[0]) #wh [0,1,....wh-1]
        coords_w = torch.arange(self.window_size[1]) #ww [0,1,....ww-1]
        #stack合并tensor
        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
        #将coords按coords[1]这一维度将此维度及之后的维度合并成一维的
        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
        #None可以看作给该维度加上一个[] 如第一个[:,:,None]就是给最内维的维度中所有元素加上[]
        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)
        #
        #dim -> 3*dim (q k v)
        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)
        #用截断正态分布填充position bias矩阵
        trunc_normal_(self.relative_position_bias_table, std=.02)
        self.softmax = nn.Softmax(dim=-1)

    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
        """
        # batch*num_windows wh*ww c-dimension
        B_, N, C = x.shape
        #batch*num_windows wh*ww c 经过qkv线性变换为 num_windows*batch,wh*ww,3*c  3表示q k v三个矩阵
        # batch*num_windows wh*ww 3 num_heads c-dimention -> 3 ,num_windows*batch, num_heads ,wh*ww ,c//num_heads
        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)    num_windows*batch ,num_head ,wh*ww ,c//num_heads

        q = q * self.scale  #num_windows*batch num_head wh*ww c//num_heads
        #计算对应位置元素相乘在求和
        attn = (q @ k.transpose(-2, -1)) # num_windows*batch,num_head ,wh*ww, wh*ww

        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()  # num_head, Wh*Ww, Wh*Ww
        #unsqueeze用于增加一个维度 等同[None,:,:]
        attn = attn + relative_position_bias.unsqueeze(0) # num_windows*batch,num_head ,wh*ww, wh*ww
        #是否进行mask
        if mask is not None:
            nW = mask.shape[0] #num_mask
            #attn + mask
            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)

        attn = self.attn_drop(attn)
        #num_windows*batch,num_head ,wh*ww, c//num_heads-> batch*num_windows, wh*ww ,c-dimension
        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x

 下面把attention相关位置编码的代码拿出来详细讲解一下

以下部分主要选自于图解swin transformer

对于attention来说，以不同元素为原点，其他元素的坐标也是不同的，以window_size=2为例，其相对位置编码如下图

源代码

coords_h = torch.arange(self.window_size[0]) #wh [0,1,....wh-1]
coords_w = torch.arange(self.window_size[1]) #ww [0,1,....ww-1]
#stack合并tensor
coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
"""
经过meshgrid后
  (tensor([[0, 0],
           [1, 1]]), 
   tensor([[0, 1],
           [0, 1]]))
经过stack后
tensor([[[0, 0],
         [1, 1]],

        [[0, 1],
         [0, 1]]])
"""
#将coords按coords[1]这一维度将此维度及之后的维度合并成一维的
coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
"""
经过flatten后
tensor([[0, 0, 1, 1],
        [0, 1, 0, 1]])
"""

#None可以看作给该维度加上一个[] 如第一个[:,:,None]就是给最内维的维度中所有元素加上[]
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
#注册一个不参与网络学习的变量
self.register_buffer("relative_position_index", relative_position_index)

 window_size为2的例子

>>> import torch
>>> window_size = [2,2]
>>> coords_h = torch.arange(window_size[0])
>>> coords_w = torch.arange(window_size[1])
>>> coords = torch.meshgrid([coords_h,coords_w])
>>> coords
(tensor([[0, 0],
        [1, 1]]), tensor([[0, 1],
        [0, 1]]))
>>> coords = torch.stack(coords)
>>> coords
tensor([[[0, 0],
         [1, 1]],

        [[0, 1],
         [0, 1]]])
>>> coords_flatten = torch.flatten(coords,1)
>>> coords_flatten
tensor([[0, 0, 1, 1],
        [0, 1, 0, 1]])
>>> relative_coords_first = coords_flatten[:,:,None]
>>> relative_coords_second = coords_flatten[:,None,:]
>>> relative_coords = relative_coords_first - relative_coords_second
>>> relative_coords
tensor([[[ 0,  0, -1, -1],
         [ 0,  0, -1, -1],
         [ 1,  1,  0,  0],
         [ 1,  1,  0,  0]],

        [[ 0, -1,  0, -1],
         [ 1,  0,  1,  0],
         [ 0, -1,  0, -1],
         [ 1,  0,  1,  0]]])
>>> relative_coords = relative_coords.permute(1, 2, 0).contiguous()
>>> relative_coords
tensor([[[ 0,  0],
         [ 0, -1],
         [-1,  0],
         [-1, -1]],

        [[ 0,  1],
         [ 0,  0],
         [-1,  1],
         [-1,  0]],

        [[ 1,  0],
         [ 1, -1],
         [ 0,  0],
         [ 0, -1]],

        [[ 1,  1],
         [ 1,  0],
         [ 0,  1],
         [ 0,  0]]])
>>> relative_coords[:,:,0] += window_size[0] - 1
>>> relative_coords[:,:,1] += window_size[1] - 1
>>> relative_coords
tensor([[[1, 1],
         [1, 0],
         [0, 1],
         [0, 0]],

        [[1, 2],
         [1, 1],
         [0, 2],
         [0, 1]],

        [[2, 1],
         [2, 0],
         [1, 1],
         [1, 0]],

        [[2, 2],
         [2, 1],
         [1, 2],
         [1, 1]]])

可以看到上述代码最后得到的tensor就是图中最后得到的矩阵

后续我们需要将二维坐标转化为一维坐标，如果只是通过将（x,y）中两坐标相加转换为一维坐标，则对于（1，2）和（2，1）这两个二维坐标映射到一维坐标上就会是相等的，都为3。

所以，为了使得类似（1，2）和（2，1）这两个坐标在映射后的一维坐标不同，我们还需要做一些操作，如下代码

>>> relative_coords[:,:,0] *= 2 * window_size[1] - 1
>>> relative_coords
tensor([[[3, 1],
         [3, 0],
         [0, 1],
         [0, 0]],

        [[3, 2],
         [3, 1],
         [0, 2],
         [0, 1]],

        [[6, 1],
         [6, 0],
         [3, 1],
         [3, 0]],

        [[6, 2],
         [6, 1],
         [3, 2],
         [3, 1]]])
>>> relative_position_index = relative_coords.sum(-1)
>>> relative_position_index
tensor([[4, 3, 1, 0],
        [5, 4, 2, 1],
        [7, 6, 4, 3],
        [8, 7, 5, 4]])

Shifted Window Attention

使用原因：The window-based self-attention module lacks connections across windows,which limits its modeling power.To introduce cross-window connections while maintaining the efficient computation of non-overlapping windows,we propose a shifted window partitioning approach which alternates between two partitioning configuration in consecutive Swin Transformer blocks.

如上图，layer l为一个输入attention中的输入，红色框为windows,灰色框为一个patch，layer l中的分割线经过上移两个patch和左移两个patch便得到了layer l+1。原来输入可以被分割为4个windows，而经过shifted后，输入被分割为了3*3个一样大小的windows。

论文中给出了两种计算layerl +1中attention的方法：

A naive solution is to pad the smaller windows to a size of M*M and mask out the padded values when computing attention.

但上述方法会增加计算复杂度，所以给出了一个更加有效的方法

Here,we propose a more efficient batch computation approach by cyclic-shifting toward the top-left direction,as illustrated in Figure 4,the next figure.

在实际代码里，我们是通过对特征图移位，并给Attention设置mask来简介实现的，能在保持原有的window的数量下，获得等价的结果。

特征图移位操作

代码里对特征图移位是通过torch.roll()来实现的，如下图

注意：如果需要reverse cyclic shift的话只需把参数shifts设置为对应的正数值

Attention Mask

通过设置合理的mask，让shifted window attention 在与window attention相同的窗口个数下，达到等价的计算结果。

首先对shift window后的每个窗口都给上index，并做一个roll操作（window_size=2,shift_size=1)

我们希望在计算Attention的时候，让具有相同index qk进行计算，而忽略不同index qk的计算结果。

正确结果如下图所示

而想要在原始四个窗口下得到正确的结果，我们就必须要给attention的结果加入一个mask（如上图最右边所示）

        if self.shift_size > 0:
            # 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

            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)
            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)

 通过如上代码，我们会得到这样一个mask

tensor([[[[[   0.,    0.,    0.,    0.],
           [   0.,    0.,    0.,    0.],
           [   0.,    0.,    0.,    0.],
           [   0.,    0.,    0.,    0.]]],


         [[[   0., -100.,    0., -100.],
           [-100.,    0., -100.,    0.],
           [   0., -100.,    0., -100.],
           [-100.,    0., -100.,    0.]]],


         [[[   0.,    0., -100., -100.],
           [   0.,    0., -100., -100.],
           [-100., -100.,    0.,    0.],
           [-100., -100.,    0.,    0.]]],


         [[[   0., -100., -100., -100.],
           [-100.,    0., -100., -100.],
           [-100., -100.,    0., -100.],
           [-100., -100., -100.,    0.]]]]])

在之前的window attention模块的前向代码里，包含这么一段

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)

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

Swin Transformer Blocks CODE

class SwinTransformerBlock(nn.Module):
    r""" Swin Transformer Block.

    Args:
        dim (int): Number of input channels. c-dimention
        input_resolution (tuple[int]): Input resulotion. wh*ww
        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表示self attention的输入之间的位置信息或说位置关系
        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
        #判断输入张量大小与窗口的大小
        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"
        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)

        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        #隐藏层维度
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
        #shifting windows
        #mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
        if self.shift_size > 0:
            # 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

            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)
            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)

    def forward(self, x):
        #获取输入feature map的高和宽
        H, W = self.input_resolution
        #获得整个输入x batch l=h*w c-dimention
        B, L, C = x.shape
        assert L == H * W, "input feature has wrong size"

        shortcut = x
        #layer normalization 标准化
        x = self.norm1(x)
        x = x.view(B, H, W, C)

        # cyclic shift torch.roll用来shift元素
        if self.shift_size > 0:
            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
        else:
            shifted_x = x

        # 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

        # W-MSA/SW-MSA
        attn_windows = self.attn(x_windows, mask=self.attn_mask)  # nW*B, window_size*window_size, C

        # 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

        # 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)

        # FFN Feed-Forward Network #residual connection
        x = shortcut + self.drop_path(x) 
        # MLP #residual connection
        x = x + self.drop_path(self.mlp(self.norm2(x))) 

        return x

    def extra_repr(self) -> str:
        return f"dim={self.dim}, input_resolution={self.input_resolution}, num_heads={self.num_heads}, " \\
               f"window_size={self.window_size}, shift_size={self.shift_size}, mlp_ratio={self.mlp_ratio}"

    def flops(self):
        flops = 0
        H, W = self.input_resolution
        # norm1
        flops += self.dim * H * W
        # W-MSA/SW-MSA
        nW = H * W / self.window_size / self.window_size
        flops += nW * self.attn.flops(self.window_size * self.window_size)
        # mlp
        flops += 2 * H * W * self.dim * self.dim * self.mlp_ratio
        # norm2
        flops += self.dim * H * W
        return flops

BasicLayer

如图中，每个stage中都有偶数个swin transformer blocks，右边的图中为两个相连的blocks的结构，下面代码给出了有depth个block的stage的实现模型

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

        # build blocks
        #nn.modulelist: Holds submodules in a list.
        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)])

        # patch merging layer
        if downsample is not None:
            self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer)
        else:
            self.downsample = None

    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

    def extra_repr(self) -> str:
        return f"dim={self.dim}, input_resolution={self.input_resolution}, depth={self.depth}"

    def flops(self):
        flops = 0
        for blk in self.blocks:
            flops += blk.flops()
        if self.downsample is not None:
            flops += self.downsample.flops()
        return flops

SwinTransformer

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

        # 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)

        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

        # 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)

        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)

    @torch.jit.ignore
    def no_weight_decay(self):
        return {'absolute_pos_embed'}

    @torch.jit.ignore
    def no_weight_decay_keywords(self):
        return {'relative_position_bias_table'}

    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

    def flops(self):
        flops = 0
        flops += self.patch_embed.flops()
        for i, layer in enumerate(self.layers):
            flops += layer.flops()
        flops += self.num_features * self.patches_resolution[0] * self.patches_resolution[1] // (2 ** self.num_layers)
        flops += self.num_features * self.num_classes
        return flops