小白理解SwinTransformer代码，【掰开揉碎】逐条理解！！

最新推荐文章于 2024-05-03 16:35:29 发布

A man protect you

最新推荐文章于 2024-05-03 16:35:29 发布

阅读量176

点赞数

文章标签：计算机视觉 transformer 图像处理 pytorch

本文链接：https://blog.csdn.net/weixin_46083369/article/details/134379229

版权

Swin Transformer功能实现与原理分析

前言
Swin Transformer
操作1、PatchEmbed
操作2、BasicLayer

前言

写的很基础，因为刚入门图像分类，pytorch也不是很熟练。所以讲得有点基础，本意是供自己使用的。文中借鉴了许多其他博主的见解，有的标明了链接，有的借鉴了文中观点没有标明出处。请多多包涵！

Swin Transformer

Swin Transformer的作用是模型的参数预设定、图片在进入后续Layers的预处理**(Patch Partition和Linear Embedding)**和按照不同的层序创建不同的BasicLayer (Swin Transformer Block和Patch Merging)。

class SwinTransformer(nn.Module):
    r""" 
    Swin Transformer
    """

    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, fused_window_process=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，根据层序来设置不同层中的不同配置参数！
        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,
                               fused_window_process=fused_window_process)
            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 forward_features(self, x):
        # 操作一：patch ”划分“ 和 ”嵌入“，两个函数看作一个整体，PatchEmbed
        x = self.patch_embed(x)
        if self.ape:
            x = x + self.absolute_pos_embed
        x = self.pos_drop(x)  # 对位置编码进行随机失活

        # 操作二：进入每一stage执行， BasicLayer
        for layer in self.layers:
            x = layer(x)

        # 操作三：进行图像分类操作
        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):
        # 1、进行Swin Transformer整体架构的搭建和为分类任务做准备
        x = self.forward_features(x)
        # 2、进行分类
        x = self.head(x)
        return x

对init()和forward()的一些感悟:——为啥init()中参数这么多？还都有默认初始值？

可以理解为在模型训练或测试之前，即模型中的各层输入信息、模块配置、输出信息都是在init()中按照参数中给定的标准来准备好的，forward()则在应用时将各个模块进行组装，然后引入真实输入，产生真实输出！
可以看见即使默认参数和实参部分相同，也会再写一遍。应该是为了程序一目了然吧！

主要操作都在forward_features和forward中：

操作1：patch ”划分“ 和 ”嵌入“，两个函数看作一个整体，PatchEmbed
操作2：进入每一自定义stage执行，其相应的参数都一一标明， BasicLayer
操作3：进行图像分类操作

操作1、PatchEmbed

class PatchEmbed(nn.Module):
    r""" 
    Image to Patch Embedding
    """
    def __init__(self, img_size=224, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
        super().__init__()
        to_2tuple()
        # 函数的作用是将一个值转换为长度为2的元组。
        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

        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

    def forward(self, x):
        B, C, H, W = x.shape
        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]})."
        
        # 唯一操作：将RGB三通道图像按照4×4分成一个个patches，同时得到指定维度的嵌入，然后再将每一个patch长和宽展平
        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

-------------------------------------------疑问：为啥这里的x：（B，C，W，H）？？-----------------------------
解释： 为了符合torch.nn.Conv2d(）的输入形式！！

在这里插入图片描述 ----- 1、self.proj(x): 用卷积对x做下采样patch_size倍**（B，embed_dim，W/4，H/4）** 。这里并不是想象中那样——将原图中的4×4像素看作一个patch，而是将这个patch的三层做卷积，然后再做embed_dim次不同卷积 2、.flatten(2): 可以接受一个参数start_dim，指定从哪一维开始展平，默认为0（即从第一维开始展平），即将上一步的特征图张量从第2个维度开始展开，将后面的维度转化为一维**(B，C，W/4 * H/4)。 3、.transpose(1, 2): 最后，对上一步得到的二维张量进行转置操作，将第1维和第2维进行交换(B，W/4 * H/4，C)**。

操作2、BasicLayer

class BasicLayer(nn.Module):
    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,
                 fused_window_process=False):

        super().__init__()
        self.dim = dim
        self.input_resolution = input_resolution
        self.depth = depth
        self.use_checkpoint = use_checkpoint

        # 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,
                                 fused_window_process=fused_window_process)
            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):
    	# 操作一：按照Swin Transformer中参数的要求，创建每一个BasicLayer的相应SwinTransformerBlock
        for blk in self.blocks:
            if self.use_checkpoint:
                x = checkpoint.checkpoint(blk, x)
            else:
                x = blk(x)
        # 操作二：根据传参来确定当前BasicLayer是否需要downsample，即PatchMerging     
        if self.downsample is not None:
            x = self.downsample(x)
        return x

操作2.1：按照Swin Transformer中参数的要求，创建每一个BasicLayer的相应SwinTransformerBlock操作2.2：根据传参来确定当前BasicLayer是否需要downsample，即PatchMerging

操作2.1、SwinTransformerBlock

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
        fused_window_process (bool, optional): If True, use one kernel to fused window shift & window partition for acceleration, similar for the reversed part. Default: False
    """

    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,
                 fused_window_process=False):
        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 # 用来控制是否需要移位，同时计算掩码

        self.mlp_ratio = mlp_ratio
		
		# if window size is larger than input resolution, we don't partition windows
        if min(self.input_resolution) <= self.window_size:
            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"

        # 与forward()中执行的步骤一一对应
        # 步骤一
        self.norm1 = norm_layer(dim)
        # 步骤二
        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)
        # 步骤三
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        # 步骤四
        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)

        # ！！关键：为移位窗口的得分矩阵做掩码矩阵！！
        if self.shift_size > 0:
            # calculate attention mask for SW-MSA
            # 1、按照不同Layer的patch特征图的尺寸建立掩码矩阵，拥有和feature map一样的通道排列顺序，方便后续window_partition
            # input_resolution = patches_resolution[0] // (2 ** i_layer)
            H, W = self.input_resolution
            img_mask = torch.zeros((1, H, W, 1))  # 1 H W 1

            # slice(0, -self.window_size)表示对一个一维数组或序列进行切片，从索引0开始，到索引-self.window_size结束（不包括-self.window_size）。
            # 具体来说，这个切片包括了从索引0开始的self.window_size个元素之前的所有元素。可以作为参数传递给序列或数组的切片操作，用于获取序列或数组的子集
            h_slices = (slice(0, -self.window_size),
                        slice(-self.window_size, -self.shift_size),
                        slice(-self.shift_size, None))  # None用于表示在切片操作中，对应维度的结束位置不限制，即直到该维度的最后一个元素为止。

            w_slices = (slice(0, -self.window_size),
                        slice(-self.window_size, -self.shift_size),
                        slice(-self.shift_size, None))

            # 根据cnt对掩码矩阵相应位置进行填值
            cnt = 0
            for h in h_slices:
                for w in w_slices:
                    img_mask[:, h, w, :] = cnt  # 按照切片对象中设定好的参数，对图像相应部分填入相应数值
                    cnt += 1

            # 将掩码矩阵进行窗口的划分
            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)

            # ！！不同维度的张量做减法！！——这个方法等效将窗口展平、装置做点乘操作（讲解的方法）效果相同！！
            # 广播操作是指在不改变张量元素数量的前提下，通过在某些维度上重复元素，使得张量可以与另一个形状不同的张量进行计算的过程。
            # 在进行广播操作时，系统会自动在较小的张量上添加新的维度或者将某些维度的大小扩展为相同的大小，以便于两个张量能够相互匹配。
            attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
            
            # ！！在 PyTorch 中，所有的张量（tensor）默认都是 float 类型！！
            # 将不为0的位置上进行掩码，即改值为-100.0，同时将为0的位置，改为0.0。为了防止精度损失
            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)
        self.fused_window_process = fused_window_process

        def forward(self, x):
        H, W = self.input_resolution
        B, L, C = x.shape # 为啥一直保留C？
        assert L == H * W, "input feature has wrong size"

        shortcut = x

        # 步骤一：LayerNorm
        x = self.norm1(x)
        x = x.view(B, H, W, C)

        # 步骤二：W-MSA\SW-MSA
        # 步骤2.1：划分窗口，window_partition
        # 判断窗口是否需要cyclic shift
        if self.shift_size > 0:
            if not self.fused_window_process:
                # ------------------------torch.roll(x, shifts, dims)------------------------
                # 将dims为看作连续可循环的整体，根据shifts的正负选择向左向右移动相应数值的位置
                shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
                # partition windows
                x_windows = window_partition(shifted_x, self.window_size)  # nW*B, window_size, window_size, C
            else:
                x_windows = WindowProcess.apply(x, B, H, W, C, -self.shift_size, self.window_size)
        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

        # 步骤2.2：对划分好窗口的张量进行自注意力计算，WindowAttention
        # 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)

        # 步骤2.3：对cyclic shift过的图像进行reverse cyclic shift，恢复原样。window_reverse
        if self.shift_size > 0:
            if not self.fused_window_process:
                shifted_x = window_reverse(attn_windows, self.window_size, H, W)  # B H' W' C
                x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
            else:
                x = WindowProcessReverse.apply(attn_windows, B, H, W, C, self.shift_size, self.window_size)
        else:
            shifted_x = window_reverse(attn_windows, self.window_size, H, W)  # B H' W' C
            x = shifted_x
        x = x.view(B, H * W, C)

        # 步骤三：与输入图像做残差连接
        x = shortcut + self.drop_path(x)

        # 步骤四：同时完成Layernorm、MLP和残差链接
        # FFN
        x = x + self.drop_path(self.mlp(self.norm2(x)))

        return x

实现SwinTransformerBlock的两种形式。
在实现W-MSA\SW-MSA的关键操作：

操作2.1.1：划分窗口，window_partition
操作2.1.2：对划分好窗口的张量进行自注意力计算，WindowAttention
操作2.1.3：对cyclic shift过的图像进行reverse cyclic shift，恢复原样。window_reverse
操作2.1.4：通过多层的线性变换和非线性激活函数对输入特征进行变换，这有助于提取更丰富的特征表示，使得模型能够更好地理解输入数据并进行更准确的预测或分类。

操作2.1.1、window_partition

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

.permute(dim0, dim1, dime2, …): 对张量进行转置，直接为当前维度序号指定为其他维度序号，则当前维度即变为其他维度。permute和不用transpose的原因，看这篇足够！！

.contiguous()： 把tensor变成在内存中连续分布的形式。看这篇足够！！

操作2.1.2、WindowAttention

WindowAttention就是在 Vision Transformer 模块的Attention基础上加入了相对位置偏移relative_position_bias_table（即论文中提出的 Relative Position Bias）来提升精度：
在这里插入图片描述
相对位置偏移讲解详见：
文章一
 文章二

# W-MSA/SW-MSA，对每个window进行自注意力，根据是否由mask来区分两种自注意力的计算，即原始Transformer中的Scaled Dot-Product Attention 和 Multi-Head Attention的结合体
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

        # 相对位置的实现！！
        # 1、随机生成相对位置偏执表
        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

        # 2、为每个窗口创建位置索引表
        coords_h = torch.arange(self.window_size[0])
        coords_w = torch.arange(self.window_size[1])
        # 将两个坐标向量进行网格化，并将其堆叠在一起
        # torch.meshgrid()函数会接受多个一维张量作为输入，并返回一个元组，其中包含了输入张量的网格化版本。例如，对于输入[1, 2, 3]和[4, 5]
        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww

        # 3、每个绝对坐标 - 其他绝对坐标，得到相对当前绝对坐标的相对坐标
        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
        # permute(1, 2, 0)会对张量进行维度转置
        # contiguous()是为了保证张量在内存中是连续存储的。
        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2

        # 4、将二维的坐标转换成一维的索引
        # 步骤一：将每一行加上window_h - 1，每一列加上window_w - 1
        relative_coords[:, :, 0] += self.window_size[0] - 1  # shift to start from 0
        relative_coords[:, :, 1] += self.window_size[1] - 1
        # 步骤二：每一行乘上2 * window_w - 1
        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
        # 步骤三：行列坐标相加，得到relative_position_index
        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
        self.register_buffer("relative_position_index", relative_position_index)

        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)

    # 按理说这里是不需要这个mask的！！
    # 因为在Swin Transformer中只适用了原始Transformer中Encoder，其中的Pad Mask在这里也用不到（图片大小相同）
    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
        # 这里将QKV三个矩阵同时计算了
        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)

        # Scaled Dot-Product Attention
        # 1、Scale
        q = q * self.scale
        # 2、MatMul
        attn = (q @ k.transpose(-2, -1))  # @符号表示矩阵乘法（矩阵点乘）

        # ！！ 这里是按照Swin Tranformer中相对位置偏执加的位置，所以在这里加入
        # self.relative_position_index中的索引信息被用来获取self.relative_position_bias_table中对应位置的相对位置偏置值，最后将其变形
        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
        # 相对位置偏执如公式中的一样与注意力权重矩阵直接相加
        attn = attn + relative_position_bias.unsqueeze(0)

        # 3/4、Mask不做！！装样子    和       SoftMax
        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)

        attn = self.attn_drop(attn)

        # 5、MatMul 和  Multi-Head Attention中的Concat
        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        
        # Multi-Head Attention中的Linear
        x = self.proj(x)
        x = self.proj_drop(x)
        return x

操作2.1.3、window_reverse

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

同window_partition

操作2.1.4、Mlp

class Mlp(nn.Module):
    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
        super().__init__()
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        self.fc1 = nn.Linear(in_features, hidden_features)
        self.act = act_layer()
        self.fc2 = nn.Linear(hidden_features, out_features)
        self.drop = nn.Dropout(drop)

    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        x = self.drop(x)
        x = self.fc2(x)
        x = self.drop(x)
        return x

操作2.2、PatchMerging

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
        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
        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."
        # 转过来
        x = x.view(B, H, W, C)
        
        # 将4×4的patch进行拆分
        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
        
        # 在通道上拼接张量
        x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
        
        # 转过去
        x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C

        x = self.norm(x)
        x = self.reduction(x)

        return x

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

    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