改进YOLOv5系列：3.Swin Transformer结构的修改

YOLOv5模块改进，适用于 YOLOv7、YOLOv5、YOLOv4、Scaled_YOLOv4、YOLOv3、YOLOR一系列YOLO算法的模块改进

一系列YOLO算法改进Trick组合！
很多Trick排列组合
助力论文
数据集涨点
创新点改进

具体博客看置顶

以下Swin_transformer 的改进

使用YOLOv5算法作为演示，模块可以无缝插入到YOLOv7、YOLOv5、YOLOv4、Scaled_YOLOv4、YOLOv3、YOLOR等一系列YOLO算法中

方法1.

YOLOv5的yaml配置文件

首先增加以下yolov5_swin_transfomrer.yaml文件

代码
# YOLOv5  by Ultralytics, GPL-3.0 license

# Parameters
nc: 80  # number of classes
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple
anchors:
  - [10,13, 16,30, 33,23]  # P3/8
  - [30,61, 62,45, 59,119]  # P4/16
  - [116,90, 156,198, 373,326]  # P5/32

# YOLOv5 -SwinV2.0 Backbone
backbone:
  # [from, number, module, args]
  [[-1, 1, PatchEmbed, [4,3,64]],  # 0-P1/2
   [-1, 1, SwinTransformer_Layer, [64,2,2]],  # 1-P2/4   [1, 128, 32, 32])
   [-1, 1, SwinTransformer_Layer, [128,2,4]],           # 2[1, 256, 16, 16])
   [-1, 1, SwinTransformer_Layer, [256,6,8,True]],                      
   [-1, 1, LayerNorm, [256]],
   [-1, 1, SwinTransformer_Layer, [256,6,8,True]],
   [-1, 1, LayerNorm, [256]],  # 3-P5/8                 
   [-1, 1, SwinTransformer_Layer, [256,6,8]],   #7    [1, 512, 8, 8])
   [-1, 1, SwinTransformer_Layer, [512,2,16,True]],  # 8  True：last_layer no PathchMerging  
   
  ]

# YOLOv5 v6.0 head
head:
  [
    [-1, 1,  nn.Conv2d, [512,256,1]],    #9      1,512,8,8
    [-1, 1, nn.Upsample, [None, 2, 'nearest']], #1,512,8,8
    [[-1, 2], 1, Concat, [1]],  # cat backbone P4
     [-1, 1, SwinTransformer_Layer, [512,2,8,True,4]],  #  [1, 512, 16, 16]) 

    [-1, 1, nn.Conv2d, [512,128,1]],    #13 [1, 256, 8, 8]
   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
    [[-1, 1], 1, Concat, [1]],  # cat backbone P3
    [-1, 1, SwinTransformer_Layer, [256,2,2,True,8]],  # 16 (P3/8-small)  torch.Size([1, 256, 32, 32])

    [-1, 1, nn.Conv2d, [256,128,1,2]],    #13  
    [[-1, 13], 1, Concat, [1]],  # cat head P4
    [-1, 1, SwinTransformer_Layer, [256,2,2,True,8]],  # 19 (P4/16-medium)     16 16

    [-1, 1, nn.Conv2d, [256,256,1,2]],  #16
    [[-1, 9], 1, Concat, [1]],  # cat head P5
    [-1, 1, SwinTransformer_Layer, [512,2,2,True,8]],  # 22 (P5/32-large)

   [[16,19, 22], 1, Detect, [nc, anchors,[256,256,512]]],  # Detect(P3, P4, P5)
  ]

新增或者减少模块

当需要修改yaml配置文件，将xx模块 加到你想加入的位置（层数）；
首先基于一个可以成功运行的.yaml模型配置文件，进行新增或者减少层数 之后，那么该层网络后续的层的编号都会发生改变，对应的一些层都需要针对性的修改，以匹配通道和层数的关系，
下篇文章详细讲解这块（以yolov5s.yaml文件为例，增加模块的例子）

common.py配置

在./models/common.py文件中增加以下模块，直接复制即可

# Swin_Transformer

def drop_path_f(x, drop_prob: float = 0., training: bool = False):
    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
    This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
    changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
    'survival rate' as the argument.
    """
    if drop_prob == 0. or not training:
        return x
    keep_prob = 1 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)  # work with diff dim tensors, not just 2D ConvNets
    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
    random_tensor.floor_()  # binarize
    output = x.div(keep_prob) * random_tensor
    return output


class DropPath(nn.Module):
    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks).
    """

    def __init__(self, drop_prob=None):
        super(DropPath, self).__init__()
        self.drop_prob = drop_prob

    def forward(self, x):
        return drop_path_f(x, self.drop_prob, self.training)


def window_partition(x, window_size: int):
    """
    将feature map按照window_size划分成一个个没有重叠的window
    Args:
        x: (B, H, W, C)
        window_size (int): window size(M)
    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)
    # permute: [B, H//Mh, Mh, W//Mw, Mw, C] -> [B, H//Mh, W//Mh, Mw, Mw, C]
    # view: [B, H//Mh, W//Mw, Mh, Mw, C] -> [B*num_windows, Mh, Mw, C]
    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
    return windows


def window_reverse(windows, window_size: int, H: int, W: int):
    """
    将window还原成一个feature map
    Args:
        windows: (num_windows*B, window_size, window_size, C)
        window_size (int): Window size(M)
        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))
    # view: [B*num_windows, Mh, Mw, C] -> [B, H//Mh, W//Mw, Mh, Mw, C]
    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
    # permute: [B, H//Mh, W//Mw, Mh, Mw, C] -> [B, H//Mh, Mh, W//Mw, Mw, C]
    # view: [B, H//Mh, Mh, W//Mw, Mw, C] -> [B, H, W, C]
    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
    return x


class Mlp(nn.Module):
    """ MLP as used in Vision Transformer, MLP-Mixer and related networks
    """

    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.drop1 = nn.Dropout(drop)
        self.fc2 = nn.Linear(hidden_features, out_features)
        self.drop2 = nn.Dropout(drop)

    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        x = self.drop1(x)
        x = self.fc2(x)
        x = self.drop2(x)
        return x


class LayerNorm(nn.Module):
    """ MLP as used in Vision Transformer, MLP-Mixer and related networks
    """

    def __init__(self, dim):
        super().__init__()
        self.norm = nn.LayerNorm(dim)

    def forward(self, x):
        # print("0:",x.size())
        x = x.permute(0, 3, 2, 1).contiguous()  # B,H,W,C
        x = self.norm(x)
        # print("1:",x.size())
        return x.permute(0, 3, 2, 1).contiguous()


class PatchEmbed(nn.Module):
    """ Image to Patch Embedding

    Args:
        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, patch_size=4, in_chans=3, embed_dim=96, norm_layer=nn.LayerNorm):
        super().__init__()
        patch_size = (patch_size, patch_size)
        self.patch_size = patch_size

        self.in_chans = in_chans
        self.embed_dim = embed_dim

        # liner embedding --->conv
        self.tokens = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)

        # LayerNorm
        if norm_layer is not None:
            self.norm = norm_layer(embed_dim)
        else:
            self.norm = None
        self.pos_drop = nn.Dropout(p=0.)

    def forward(self, x):
        """Forward function."""
        # 在下方或者是右侧进行padding  保证patchsize整除
        _, _, H, W = x.size()

        # print("pasize:",self.patch_size)
        if W % self.patch_size[1] != 0:
            x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1]))
        if H % self.patch_size[0] != 0:
            x = F.pad(x, (0, 0, 0, self.patch_size[0] - H % self.patch_size[0]))
        # wh * Ww  tokens
        # print("下采样倍数：",self.patch_size)
        x = self.tokens(x)  # B C Wh Ww
        # print('self.norm',self.norm)
        _, _, H, W = x.shape
        # flatten: [B, C, H, W] -> [B, C, HW]
        # transpose: [B, C, HW] -> [B, HW, C]
        if self.norm is not None:
            Wh, Ww = x.size(2), x.size(3)
            x = x.flatten(2).transpose(1, 2)

            x = self.norm(x)
            x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww)
        x = self.pos_drop(x)
        # print('PatchEMbed output shape:',x.size())
        return x


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

    Args:
        dim (int): Number of input channels. 
        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
    """

    def __init__(self, dim, norm_layer=nn.LayerNorm):
        super().__init__()
        self.dim = dim
        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
        self.norm = norm_layer(4 * dim)

    def forward(self, x, H, W):
        """ Forward function.

        Args:
            x: Input feature, tensor size (B, H*W, C).
            H, W: Spatial resolution of the input feature.
        """

        B, C, H, W = x.shape
        # print('------------------------PatchMErging input shape:',x.size())
        # H=L**0.5
        # W=H
        # assert L == H * W, "input feature has wrong size"
        # assert H==W , "input feature has wrong size"
        x = x.view(B, int(H), int(W), C)

        # padding
        pad_input = (H % 2 == 1) or (W % 2 == 1)
        if pad_input:
            x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2))

        x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C 左上
        x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C 左下 
        x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C 右上
        x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C 右下
        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)  # B H/2*W/2 2*C 
        # print('PatchMerging output shape:',x.size())
        return x


class Global_WindowAttention(nn.Module):
    r""" MOA - multi-head self attention (W-MSA) module with relative position bias.
    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, input_resolution, 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.query_size = self.window_size[0]

        self.key_size = self.window_size[0] + 2
        h, w = input_resolution
        self.seq_len = h // self.query_size

        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5
        self.reduction = 32
        self.pre_conv = nn.Conv2d(dim, int(dim // self.reduction), 1)
        # define a parameter table of relative position bias
        self.relative_position_bias_weight = nn.Parameter(
            torch.zeros((2 * self.seq_len - 1) * (2 * self.seq_len - 1), num_heads))  # 2*Wh-1 * 2*Ww-1, nH
        # print(self.relative_position_bias_weight.shape)
        # get pair-wise relative position index for each token inside the window
        coords_h = torch.arange(self.seq_len)
        coords_w = torch.arange(self.seq_len)
        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.seq_len - 1  # shift to start from 0
        relative_coords[:, :, 1] += self.seq_len - 1
        relative_coords[:, :, 0] *= 2 * self.seq_len - 1
        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww

        self.register_buffer("relative_position_index", relative_position_index)
        self.queryembedding = Rearrange('b c (h p1) (w p2) -> b (p1 p2 c) h w', p1=self.query_size, p2=self.query_size)
        self.keyembedding = nn.Unfold(kernel_size=(self.key_size, self.key_size), stride=14, padding=1)
        self.query_dim = int(dim // self.reduction) * self.query_size * self.query_size
        self.key_dim = int(dim // self.reduction) * self.key_size * self.key_size

        self.q = nn.Linear(self.query_dim, self.dim, bias=qkv_bias)
        self.kv = nn.Linear(self.key_dim, 2 * self.dim, 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_weight, std=.02)
        self.softmax = nn.Softmax(dim=-1)

    def forward(self, x, H, W):
        """
        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, H, W, C = x.shape
        B, _, C = x.shape
        x = x.reshape(-1, C, H, W)
        x = self.pre_conv(x)
        query = self.queryembedding(x).view(B, -1, self.query_dim)
        query = self.q(query)
        B, N, C = query.size()
        q = query.reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)
        key = self.keyembedding(x).view(B, -1, self.key_dim)
        kv = self.kv(key).reshape(B, N, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        k = kv[0]
        v = kv[1]
        q = q * self.scale
        attn = (q @ k.transpose(-2, -1))
        relative_position_bias = self.relative_position_bias_weight[self.relative_position_index.view(-1)].view(
            self.seq_len * self.seq_len, self.seq_len * self.seq_len, -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)
        attn = self.softmax(attn)
        attn = self.attn_drop(attn)
        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


class WindowAttention(nn.Module):
    """ 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
        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, attn_drop=0., proj_drop=0.,
                 meta_network_hidden_features=256):

        super().__init__()
        self.dim = dim
        self.window_size = window_size  # [Mh, Mw]
        self.num_heads = num_heads
        head_dim = dim // num_heads
        # self.scale = head_dim ** -0.5

        # define a parameter table of relative position bias
        self.relative_position_bias_weight = nn.Parameter(
            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads))  # [2*Mh-1 * 2*Mw-1, nH]
        # 获取窗口内每对token的相对位置索引
        # 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, Mh, Mw]indexing="ij"
        # coords_flatten = torch.flatten(coords, 1)  # [2, Mh*Mw]
        # # [2, Mh*Mw, 1] - [2, 1, Mh*Mw]
        # relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # [2, Mh*Mw, Mh*Mw]
        # relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # [Mh*Mw, Mh*Mw, 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)  # [Mh*Mw, Mh*Mw]
        # self.register_buffer("relative_position_index", relative_position_index)
        # Init meta network for positional encodings
        self.meta_network: nn.Module = nn.Sequential(
            nn.Linear(in_features=2, out_features=meta_network_hidden_features, bias=True),
            nn.ReLU(inplace=True),
            nn.Linear(in_features=meta_network_hidden_features, out_features=num_heads, bias=True))
        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)

        nn.init.trunc_normal_(self.relative_position_bias_weight, std=.02)
        self.softmax = nn.Softmax(dim=-1)
        # Init tau
        self.register_parameter("tau", nn.Parameter(torch.zeros(1, num_heads, 1, 1)))

        # Init pair-wise relative positions (log-spaced)
        indexes = torch.arange(self.window_size[0], device=self.tau.device)
        coordinates = torch.stack(torch.meshgrid([indexes, indexes]), dim=0)
        coordinates = torch.flatten(coordinates, start_dim=1)
        relative_coordinates = coordinates[:, :, None] - coordinates[:, None, :]
        relative_coordinates = relative_coordinates.permute(1, 2, 0).reshape(-1, 2).float()
        relative_coordinates_log = torch.sign(relative_coordinates) \
                                   * torch.log(1. + relative_coordinates.abs())
        self.register_buffer("relative_coordinates_log", relative_coordinates_log)

    def get_relative_positional_encodings(self):
        """
        Method computes the relative positional encodings
        :return: Relative positional encodings [1, number of heads, window size ** 2, window size ** 2]
        """
        relative_position_bias = self.meta_network(self.relative_coordinates_log)
        relative_position_bias = relative_position_bias.permute(1, 0)
        relative_position_bias = relative_position_bias.reshape(self.num_heads,
                                                                self.window_size[0] * self.window_size[1], \
                                                                self.window_size[0] * self.window_size[1])
        return relative_position_bias.unsqueeze(0)

    def forward(self, x, mask=None):
        """
        Args:
            x: input features with shape of (num_windows*B, Mh*Mw, C)
            mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
        """
        # [batch_size*num_windows, Mh*Mw, total_embed_dim]
        B_, N, C = x.shape
        # qkv(): -> [batch_size*num_windows, Mh*Mw, 3 * total_embed_dim]
        # reshape: -> [batch_size*num_windows, Mh*Mw, 3, num_heads, embed_dim_per_head]
        # permute: -> [3, batch_size*num_windows, num_heads, Mh*Mw, embed_dim_per_head]
        qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        # [batch_size*num_windows, num_heads, Mh*Mw, embed_dim_per_head]
        q, k, v = qkv.unbind(0)  # make torchscript happy (cannot use tensor as tuple)
        attn = torch.einsum("bhqd, bhkd -> bhqk", q, k) \
               / torch.maximum(torch.norm(q, dim=-1, keepdim=True)
                               * torch.norm(k, dim=-1, keepdim=True).transpose(-2, -1),
                               torch.tensor(1e-06, device=q.device, dtype=q.dtype))
        # transpose: -> [batch_size*num_windows, num_heads, embed_dim_per_head, Mh*Mw]
        # @: multiply -> [batch_size*num_windows, num_heads, Mh*Mw, Mh*Mw]
        # q = q * self.scale
        # cosine -->dot?????? Scaled cosine attention：cosine(q,k)/tau 也许理解的不准确： 控制数值范围有利于训练稳定 （残差块的累加 导致深层难以稳定训练）
        # attn = (q @ k.transpose(-2, -1))
        # q = torch.norm(q, p=2, dim=-1)
        # k = torch.norm(k, p=2, dim=-1)
        # attn /= q.unsqueeze(-1)
        # attn /= k.unsqueeze(-2)
        # attn=attention_map
        # print('attn shape:',attn.size())
        # print('attn2 shape:',attention_map.size())
        attn /= self.tau.clamp(min=0.01)

        # relative_position_bias_table.view: [Mh*Mw*Mh*Mw,nH] -> [Mh*Mw,Mh*Mw,nH]
        # relative_position_bias = self.relative_position_bias_weight[self.relative_position_index.view(-1)].view(
        #     self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1)
        # relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # [nH, Mh*Mw, Mh*Mw]
        # print("net work new positional_enco:",self.__get_relative_positional_encodings().size())
        # print('attn shape:',attn.size())
        # attn = attn + relative_position_bias.unsqueeze(0)
        attn = attn + self.get_relative_positional_encodings()
        if mask is not None:
            # mask: [nW, Mh*Mw, Mh*Mw]
            nW = mask.shape[0]  # num_windows
            # attn.view: [batch_size, num_windows, num_heads, Mh*Mw, Mh*Mw]
            # mask.unsqueeze: [1, nW, 1, Mh*Mw, Mh*Mw]
            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)

        # @: multiply -> [batch_size*num_windows, num_heads, Mh*Mw, embed_dim_per_head]
        # transpose: -> [batch_size*num_windows, Mh*Mw, num_heads, embed_dim_per_head]
        # reshape: -> [batch_size*num_windows, Mh*Mw, total_embed_dim]
        # x = (attn @ v).transpose(1, 2).reshape(B_, N, C)  ## float()
        x = torch.einsum("bhal, bhlv -> bhav", attn, v)
        # x = self.proj(x)
        # x = self.proj_drop(x)
        # print('out shape:',x.size())
        return x


class SwinTransformerBlock(nn.Module):
    r""" Swin Transformer Block.
    Args:
        dim (int): Number of input channels.
        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
        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, num_heads, window_size=7, shift_size=0,
                 mlp_ratio=4., qkv_bias=True, drop=0., attn_drop=0., drop_path=0.,
                 act_layer=nn.GELU, norm_layer=nn.LayerNorm, Global=False):
        super().__init__()
        self.dim = dim
        self.num_heads = num_heads
        self.window_size = window_size
        self.shift_size = shift_size
        self.mlp_ratio = mlp_ratio
        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
        # patches_resolution = [img_size[0] // patch_size[0], img_size[1] // patch_size[1]]
        self.norm1 = norm_layer(dim)
        self.attn = WindowAttention(
            dim, window_size=(self.window_size, self.window_size), num_heads=num_heads, qkv_bias=qkv_bias,
            attn_drop=attn_drop, proj_drop=drop)
        # if Global else Global_WindowAttention(
        # dim, window_size=(self.window_size, self.window_size),input_resolution=() num_heads=num_heads, qkv_bias=qkv_bias,
        # 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)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, out_features=dim, act_layer=act_layer,
                       drop=drop)

    def forward(self, x, attn_mask):
        # H, W = self.H, self.W
        # print("org-input block shape:",x.size())
        x = x.permute(0, 3, 2, 1).contiguous()  # B,H,W,C

        B, H, W, C = x.shape
        # B, L, C = x.shape
        # assert L == H * W, "input feature has wrong size"
        shortcut = x

        # H,W=int(H), int(W)
        # x = self.norm1(x)

        # x = x.view(B, H, W, C)

        # pad feature maps to multiples of window size
        # 把feature map给pad到window size的整数倍

        # if min(H, W) < self.window_size or H % self.window_size!=0:
        # Padding = True
        pad_l = pad_t = 0
        pad_r = (self.window_size - W % self.window_size) % self.window_size
        pad_b = (self.window_size - H % self.window_size) % self.window_size
        x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b))
        _, Hp, Wp, _ = x.shape

        # 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
            attn_mask = None

        # partition windows
        x_windows = window_partition(shifted_x, self.window_size)  # [nW*B, Mh, Mw, C]
        x_windows = x_windows.view(-1, self.window_size * self.window_size, C)  # [nW*B, Mh*Mw, C]

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

        # merge windows
        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)  # [nW*B, Mh, Mw, C]
        shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp)  # [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

        if pad_r > 0 or pad_b > 0:
            # 把前面pad的数据移除掉
            x = x[:, :H, :W, :].contiguous()
        x = self.norm1(x)  # pos-norm.1
        # x = x.view(B, H * W, C)

        # FFN
        x = shortcut + self.drop_path(x)
        x = x + self.drop_path(self.norm2(self.mlp(x)))  # pos-norm.2

        x = x.permute(0, 3, 2, 1).contiguous()
        # print("swinblock ouput——shape：",x.size())
        return x

class SwinTransformer_Layer(nn.Module):
    """
    A basic Swin Transformer layer for one stage.
    Args:
        dim (int): Number of input channels.
        depth (int): Number of blocks.
        num_heads (int): Number of attention heads.
        window_size (int): Local window size:  7 or 8 
        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
        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, depth, num_heads, last_layer=False, window_size=7,
                 mlp_ratio=4., qkv_bias=True, drop=0., attn_drop=0.,
                 drop_path=0., norm_layer=nn.LayerNorm, downsample=PatchMerging, use_checkpoint=False):
        super().__init__()
        self.dim = dim
        self.depth = depth
        self.last_layer = last_layer
        self.window_size = window_size
        self.use_checkpoint = use_checkpoint
        self.shift_size = window_size // 2

        # build blocks
        self.blocks = nn.ModuleList([
            SwinTransformerBlock(
                dim=dim,
                num_heads=num_heads,
                window_size=window_size,
                shift_size=0 if (i % 2 == 0) else self.shift_size,
                mlp_ratio=mlp_ratio,
                qkv_bias=qkv_bias,
                drop=drop,
                attn_drop=attn_drop,
                drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
                norm_layer=norm_layer,
                Global=False)
            for i in range(depth)])

        # patch merging layer

        if self.last_layer is False:
            # print('开始进行patchmergin------打印层深度：',depth)
            self.downsample = downsample(dim=dim, norm_layer=norm_layer)
        else:
            # print('最后1层默认没有Patchmerging：',depth)
            # self.norm = norm_layer(self.num_features)
            # self.avgpool = nn.AdaptiveAvgPool1d(1)
            self.downsample = None
        self.avgpool = nn.AdaptiveAvgPool1d(1)

    def create_mask(self, x, H, W):
        # calculate attention mask for SW-MSA
        # 保证Hp和Wp是window_size的整数倍
        Hp = int(np.ceil(H / self.window_size)) * self.window_size
        Wp = int(np.ceil(W / self.window_size)) * self.window_size
        # 拥有和feature map一样的通道排列顺序，方便后续window_partition
        img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # [1, Hp, Wp, 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, Mh, Mw, 1]
        mask_windows = mask_windows.view(-1, self.window_size * self.window_size)  # [nW, Mh*Mw]
        attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)  # [nW, 1, Mh*Mw] - [nW, Mh*Mw, 1]
        # [nW, Mh*Mw, Mh*Mw]
        attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))
        return attn_mask

    def forward(self, x):
        # print('swinlayers input shape:',x.size())
        B, C, H, W = x.size()
        # H=int(L**0.5)
        # W=H
        # assert L == H * W, "input feature has wrong size"

        attn_mask = self.create_mask(x, H, W)  # [nW, Mh*Mw, Mh*Mw]
        for blk in self.blocks:
            blk.H, blk.W = H, W
            if not torch.jit.is_scripting() and self.use_checkpoint:
                x = checkpoint.checkpoint(blk, x, attn_mask)
            else:
                x = blk(x, attn_mask)
        if self.downsample is not None:
            x = self.downsample(x, H, W)
            H, W = (H + 1) // 2, (W + 1) // 2

        # if  self.last_layer:
        # x=x.view(B,H,W,C)
        # x=x.transpose(1,3)
        # x = self.norm(x)  # [B, L, C]
        # x = self.avgpool(x.transpose(1, 2))  # [B, C, 1]

        # x = x.view(B,-1,H,W)
        # x = window_reverse(x, self.window_size, H, W)  # [B, H', W', C]
        # x = torch.flatten(x, 1)
        x = x.view(B, -1, H, W)  #
        # print("Swin-Transform 层 ------------------------输出维度:",x.size())
        return x     

yolo.py配置

然后找到./models/yolo.py文件下里的parse_model函数，将类名加入进去
在 models/yolo.py文件夹下全局搜索 n = n_ = max(round(n * gd), 1) if n > 1 else n
下方按图所示 只需要增加 SwinTransformerBlock

参考示意图

训练yolov5_swin_transfomrer模型

python train.py --cfg yolov5_swin_transfomrer.yaml

针对以上yaml文件继续修改

关于yolov5_swin_transfomrer.yaml文件配置中的SwinTransformer模块，可以针对不同数据集自行再进行魔改，原理一致

理论

理论部分参考：Swin Transformer的继任者：Local Vision Transformer的革命