完整教程：swin-transformer架构解析和源码解析

swin-transformer和vit-transformer的区别是它采用了多个窗口收集信息然后进行注意力机制，然后在经过移动窗口结合信息

swin-transformer结构

swin模块和其中掩码的讲解就是Patch Partition的主导目的是将原来的大模块分成小补丁，这点和VIT类似，下面

Patch Merging讲解

位置索引（bias)讲解

模块讲解完毕，MLP在代码解析中讲解，代码太长了（600多行）我就没有像以前一样逐行讲解

# --------------------------------------------------------
# Swin Transformer
# 版权 (c) 2021 Microsoft
# 根据 MIT 许可证授权 [详见 LICENSE]
# 作者 Ze Liu
# --------------------------------------------------------
import torch
import torch.nn as nn
import torch.utils.checkpoint as checkpoint
from timm.models.layers import DropPath, to_2tuple, trunc_normal_
try:
    import os, sys
    kernel_path = os.path.abspath(os.path.join('..'))
    sys.path.append(kernel_path)
    #计算绝对路径防止出错
    from kernels.window_process.window_process import WindowProcess, WindowProcessReverse
except:
    WindowProcess = None
    WindowProcessReverse = None
    print("[Warning] Fused window process have not been installed. Please refer to get_started.md for installation.")
    #特殊处理，用try-except,如果try部分出错则切换到except部分
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
        #真值保持不变，假值则赋予or后面的值
        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)
        #ffn层，进行非线性变化，收集更多信息
    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
    #数据运行流程
def window_partition(x, window_size):
    """
    参数:
        x: (B, H, W, C)
        window_size (int): 窗口大小
    返回:
        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
#主要目的是将b和window结合，将h/window_size 得到高方面的窗口数，然后用w/window_size得到宽方面的窗口数
#将2，3维度调换，然后用contiguous()保证了数据连续性
#用view方法将前三个维度结合到一起，得到批次数乘以总的窗口数
#将窗口拆开，有利于在窗口内部进行自注意力
#反回总的窗口数
def window_reverse(windows, window_size, H, W):
    """
    参数:
        windows: (num_windows*B, window_size, window_size, C)
        window_size (int): 窗口大小
        H (int): 图像高度
        W (int): 图像宽度
    返回:
        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
#将拆分的窗口复原，合并信息
#重新回到x；(B,H,W,C)
class WindowAttention(nn.Module):
    r""" 基于窗口的多头自注意力 (W-MSA) 模块，带有相对位置偏置。
    它支持移位和非移位窗口。
    #对单个窗口进行多头注意力，可以看看我的上一期讲解
    参数:
        dim (int): 输入通道数。
        window_size (tuple[int]): 窗口的高度和宽度。
        num_heads (int): 注意力头数。
        qkv_bias (bool, optional): 如果为 True，为 query、key、value 添加可学习偏置。默认: True
        qk_scale (float | None, optional): 如果设置，覆盖默认的 head_dim ** -0.5 的 qk 缩放
        attn_drop (float, optional): 注意力权重的 dropout 比率。默认: 0.0
        proj_drop (float, optional): 输出的 dropout 比率。默认: 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#定义缩放因子
        #简单的实例化
        # 定义相对位置偏置的参数表
        self.relative_position_bias_table = nn.Parameter(
            #nn.Parameter,将位置偏置变成可学习的参数
            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)
            #设置0起点方便进行学习，总共有（2m-1）*（2m-1）个可能，所以设置这些个0矩阵，创造空间
        # 2*Wh-1 * 2*Ww-1, nH
        #看我的文章，我会细讲这一部分
        # 为窗口内每个标记获取成对相对位置索引
        coords_h = torch.arange(self.window_size[0])
        coords_w = torch.arange(self.window_size[1])
        #将位置索引的高宽表示出来
        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
        #生成一个包含所有可能的 (h, w) 组合的网格坐标，
        # 将 h 和 w 这两个 (Wh, Ww) 形状的张量堆叠在一起，沿着一个新的维度。
        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
        #将 coords 张量的第1个维度开始的部分展平
        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww
        #None,在所在维度上多加一个维度
        #将 Δh 和 Δw 放在一个张量的不同切片
        #广播机制，所有行减所有列
        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
        #调换维度并进行连续化处理
        relative_coords[:, :, 0] += self.window_size[0] - 1  # 移位以从 0 开始
        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
        #我在文章写的一系列变化
        self.register_buffer("relative_position_index", relative_position_index)
        #把编号作为模型参数的一部分
        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        #通过线性变化得到Q,K,V
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        #融合信息
        self.proj_drop = nn.Dropout(proj_drop)
        trunc_normal_(self.relative_position_bias_table, std=.02)
        #用截断正态分布生成随机数来填充你的参数
        self.softmax = nn.Softmax(dim=-1)
    def forward(self, x, mask=None):
        """
        参数:
            x: 输入特征，形状为 (num_windows*B, N, C)
            mask: (0/-inf) 掩码，形状为 (num_windows, Wh*Ww, Wh*Ww) 或 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]
        #拆分获取Q,K,V
        q = q * self.scale
        attn = (q @ k.transpose(-2, -1))
        #Q*K的装置，计算注意力分数
        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)
        #将index用table表示，然后回复原来的的状态
        # 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)
        #对应公式中的+b
        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)
            #进行转换，使两者处于相同维度
            attn = attn.view(-1, self.num_heads, N, N)
            attn = self.softmax(attn)
            #softmax处理
        else:
            attn = self.softmax(attn)
        attn = self.attn_drop(attn)
        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        #对应公式中的*v
        x = self.proj(x)
        x = self.proj_drop(x)
        return x
    #拼接并进性线性变化
    def extra_repr(self) -> str:
        return f'dim={self.dim}, window_size={self.window_size}, num_heads={self.num_heads}'
    def flops(self, N):
        # 计算 1 个窗口中标记长度为 N 的 flops
        flops = 0
        # qkv = self.qkv(x)
        flops += N * self.dim * 3 * self.dim
        # attn = (q @ k.transpose(-2, -1))
        flops += self.num_heads * N * (self.dim // self.num_heads) * N
        #  x = (attn @ v)
        flops += self.num_heads * N * N * (self.dim // self.num_heads)
        # x = self.proj(x)
        flops += N * self.dim * self.dim
        return flops
    #计算浮点数的，不重要
class SwinTransformerBlock(nn.Module):
    r""" Swin Transformer 块。
    参数:
        dim (int): 输入通道数。
        input_resolution (tuple[int]): 输入分辨率。
        num_heads (int): 注意力头数。
        window_size (int): 窗口大小。
        shift_size (int): SW-MSA 的移位大小。
        mlp_ratio (float): mlp 隐藏维度与嵌入维度的比率。
        qkv_bias (bool, optional): 如果为 True，为 query、key、value 添加可学习偏置。默认: True
        qk_scale (float | None, optional): 如果设置，覆盖默认的 head_dim ** -0.5 的 qk 缩放。
        drop (float, optional): Dropout 率。默认: 0.0
        attn_drop (float, optional): 注意力 dropout 率。默认: 0.0
        drop_path (float, optional): 随机深度率。默认: 0.0
        act_layer (nn.Module, optional): 激活层。默认: nn.GELU
        norm_layer (nn.Module, optional): 归一化层。默认: nn.LayerNorm
        fused_window_process (bool, optional): 如果为 True，使用一个内核融合窗口移位和窗口分区以加速，逆过程类似。默认: 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
        self.mlp_ratio = mlp_ratio
        #实例化
        if min(self.input_resolution) <= self.window_size:
            # 如果窗口大小大于输入分辨率，则不分区窗口
            self.shift_size = 0
            self.window_size = min(self.input_resolution)
        assert 0 <= self.shift_size < self.window_size, "shift_size 必须在 0-window_size 之间"
        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)
        mlp_hidden_dim = int(dim * mlp_ratio)
        #ffn同理
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
        if self.shift_size > 0:
            # 计算 SW-MSA 的注意力掩码
            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))
            #切割图片
            cnt = 0
            for h in h_slices:
                for w in w_slices:
                    img_mask[:, h, w, :] = cnt
                    cnt += 1
                    #给切割后的图片标号,相同的数字是相邻模块
            mask_windows = window_partition(img_mask, self.window_size)
            #上文的window_partition切割图片
            # nW, window_size, window_size, 1
            mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
            #nw,window_size*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
            #掩码处理
        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
        assert L == H * W, "输入特征大小错误"
        shortcut = x
        x = self.norm1(x)
        x = x.view(B, H, W, C)
        # 循环移位
        if self.shift_size > 0:
            if not self.fused_window_process:
                shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
                # 分区窗口
                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
            # 分区窗口
            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
        # 合并窗口
        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
        # 逆循环移位
        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)
        x = shortcut + self.drop_path(x)
        # FFN
        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
class PatchMerging(nn.Module):
    r""" 补丁合并层。
    参数:
        input_resolution (tuple[int]): 输入特征的分辨率。
        dim (int): 输入通道数。
        norm_layer (nn.Module, optional): 归一化层。默认: 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, "输入特征大小错误"
        assert H % 2 == 0 and W % 2 == 0, f"x 大小 ({H}*{W}) 不是偶数。"
        x = x.view(B, H, W, C)
        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
        #在我的文章里面讲了
        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
class BasicLayer(nn.Module):
    """ Swin Transformer 的一个基本层，用于一个阶段。
    参数:
        dim (int): 输入通道数。
        input_resolution (tuple[int]): 输入分辨率。
        depth (int): 块的数量。
        num_heads (int): 注意力头数。
        window_size (int): 局部窗口大小。
        mlp_ratio (float): mlp 隐藏维度与嵌入维度的比率。
        qkv_bias (bool, optional): 如果为 True，为 query、key、value 添加可学习偏置。默认: True
        qk_scale (float | None, optional): 如果设置，覆盖默认的 head_dim ** -0.5 的 qk 缩放。
        drop (float, optional): Dropout 率。默认: 0.0
        attn_drop (float, optional): 注意力 dropout 率。默认: 0.0
        drop_path (float | tuple[float], optional): 随机深度率。默认: 0.0
        norm_layer (nn.Module, optional): 归一化层。默认: nn.LayerNorm
        downsample (nn.Module | None, optional): 层末端的下采样层。默认: None
        use_checkpoint (bool): 是否使用检查点以节省内存。默认: False
        fused_window_process (bool, optional): 如果为 True，使用一个内核融合窗口移位和窗口分区以加速，逆过程类似。默认: 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,
                 fused_window_process=False):
        super().__init__()
        self.dim = dim
        self.input_resolution = input_resolution
        self.depth = depth
        self.use_checkpoint = use_checkpoint
        # 构建块
        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)])
        # 补丁合并层
        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
class PatchEmbed(nn.Module):
    r""" 图像到补丁嵌入
    参数:
        img_size (int): 图像大小。默认: 224。
        patch_size (int): 补丁标记大小。默认: 4。
        in_chans (int): 输入图像通道数。默认: 3。
        embed_dim (int): 线性投影输出通道数。默认: 96。
        norm_layer (nn.Module, optional): 归一化层。默认: 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)
       #输入转换为一个包含两个元素的元组
        patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]]
        #长宽方向的补丁数目
        self.img_size = img_size
        self.patch_size = patch_size
        self.patches_resolution = patches_resolution
        self.num_patches = patches_resolution[0] * patches_resolution[1]
        #patches_resolution 用于跟踪特征图的分辨率
        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)
        #进行下采样
        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
        # FIXME 考虑放宽大小约束
        assert H == self.img_size[0] and W == self.img_size[1], \
            f"输入图像大小 ({H}*{W}) 与模型不匹配 ({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
    #若想进行多尺度训练，则用pading处理
    def flops(self):
        Ho, Wo = self.patches_resolution
        flops = Ho * Wo * self.embed_dim * self.in_chans * (self.patch_size[0] * self.patch_size[1])
        if self.norm is not None:
            flops += Ho * Wo * self.embed_dim
        return flops
class SwinTransformer(nn.Module):
    r""" Swin Transformer
        PyTorch 实现：`Swin Transformer: Hierarchical Vision Transformer using Shifted Windows` -
          https://arxiv.org/pdf/2103.14030
    参数:
        img_size (int | tuple(int)): 输入图像大小。默认 224
        patch_size (int | tuple(int)): 补丁大小。默认: 4
        in_chans (int): 输入图像通道数。默认: 3
        num_classes (int): 分类头的类别数。默认: 1000
        embed_dim (int): 补丁嵌入维度。默认: 96
        depths (tuple(int)): 每个 Swin Transformer 层的深度。
        num_heads (tuple(int)): 不同层中的注意力头数。
        window_size (int): 窗口大小。默认: 7
        mlp_ratio (float): mlp 隐藏维度与嵌入维度的比率。默认: 4
        qkv_bias (bool): 如果为 True，为 query、key、value 添加可学习偏置。默认: True
        qk_scale (float): 如果设置，覆盖默认的 head_dim ** -0.5 的 qk 缩放。默认: None
        drop_rate (float): Dropout 率。默认: 0
        attn_drop_rate (float): 注意力 dropout 率。默认: 0
        drop_path_rate (float): 随机深度率。默认: 0.1
        norm_layer (nn.Module): 归一化层。默认: nn.LayerNorm。
        ape (bool): 如果为 True，向补丁嵌入添加绝对位置嵌入。默认: False
        patch_norm (bool): 如果为 True，在补丁嵌入后添加归一化。默认: True
        use_checkpoint (bool): 是否使用检查点以节省内存。默认: False
        fused_window_process (bool, optional): 如果为 True，使用一个内核融合窗口移位和窗口分区以加速，逆过程类似。默认: 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, 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
        # 将图像分割成非重叠补丁
        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
        # 绝对位置嵌入
        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)
        # 随机深度
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # 随机深度衰减规则
        # 构建层
        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 _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)
        return flops

希望大家喜欢