YOLO v5加入注意力机制、swin-head、解耦头部（回归源码）

提示：文章写完后，目录可以自动生成，如何生成可参考右边的帮助文档

文章目录

YOLO v5加入注意力机制、swin-head、解耦头部（回归源码）

提示：这里可以添加本文要记录的大概内容：

在YOLO v5的backbone和head引入全局注意力机制（GAM attention）、检测头引入解耦头部、SwinTransformer_Layer层，部分commom.py代码参考地址为：

https://github.com/iloveai8086/YOLOC

提示：以下是本篇文章正文内容，下面案例可供参考

一、YOLO v5 简介

YOLO v5由输入端、Backbone、Neck、Head四部分组成。YOLO v5输入端是对图像进行预处理操作,达到对输入图片进行数据增强的效果；Backbone采用了Conv复合卷积模块、C3模块以及SPPF模块组成，Neck部分主则采用 FPN+PAN的特征金字塔结构，增加多尺度的语义表达，从而增强不同尺度上的表达能力；Head部分采用三种损失函数分别计算位置、分类和置信度损失。

二、全局注意力机制的引入

在YOLO v5 d的commom.py加入以下全局注意力机制的代码。

class GAM_Attention(nn.Module):#https://paperswithcode.com/paper/global-attention-mechanism-retain-information
    def __init__(self, c1, c2, group=True, rate=4):super(GAM_Attention, self).__init__()
        self.channel_attention = nn.Sequential(
            nn.Linear(c1,int(c1 / rate)),
            nn.ReLU(inplace=True),
            nn.Linear(int(c1 / rate), c1))
        self.spatial_attention = nn.Sequential(
            nn.Conv2d(c1, c1 // rate, kernel_size=7, padding=3, groups=rate) if group else nn.Conv2d(c1, int(c1 / rate),
                                                                                                     kernel_size=7,
                                                                                                     padding=3),
            nn.BatchNorm2d(int(c1 / rate)),
            nn.ReLU(inplace=True),
            nn.Conv2d(c1 // rate, c2, kernel_size=7, padding=3, groups=rate) if group else nn.Conv2d(int(c1 / rate), c2,
                                                                                                     kernel_size=7,
                                                                                                     padding=3),
            nn.BatchNorm2d(c2))
    def forward(self, x):
        b, c, h, w = x.shape
        x_permute = x.permute(0,2,3,1).view(b,-1, c)
        x_att_permute = self.channel_attention(x_permute).view(b, h, w, c)
        x_channel_att = x_att_permute.permute(0,3,1,2)#x_channel_att=channel_shuffle(x_channel_att,4) #last shuffle
        x = x * x_channel_att
        x_spatial_att = self.spatial_attention(x).sigmoid()
        x_spatial_att =channel_shuffle(x_spatial_att,4)  # last shuffle
        out = x * x_spatial_att
        #out=channel_shuffle(out,4) #last shufflereturn out
def channel_shuffle(x, groups=2):  ##shuffle channel
        #RESHAPE----->transpose------->Flatten
        B, C, H, W = x.size()
        out = x.view(B, groups, C // groups, H, W).permute(0, 2, 1, 3, 4).contiguous()
        out = out.view(B, C, H, W)return out

三、引入SwinTransformer_Layer层

在commom.py加入以下代码。代码较多

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 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#assertL == H * W,"input feature has wrong size"#assertH==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/24*C
        x = x.view(B,-1,4* C)  # B H/2*W/24*C
        x = self.norm(x)
        x = self.reduction(x)  # B H/2*W/22*C
        #print('PatchMerging output shape:',x.size())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 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:0.0proj_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#defineaparameter 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
        #getpair-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:0.0attn_drop(float, optional): Attention dropout rate. Default:0.0drop_path(float, optional): Stochastic depth rate. Default:0.0act_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)#ifGlobal elseGlobal_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#assertL == 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)#padfeature maps to multiples of window size
        # 把feature map给pad到window size的整数倍
        #ifmin(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
        #cyclicshiftif 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
        #partitionwindows
        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]#mergewindows
        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]#reversecyclic shiftif 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
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
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 8mlp_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:0.0attn_drop(float, optional): Attention dropout rate. Default:0.0drop_path(float| tuple[float], optional): Stochastic depth rate. Default:0.0norm_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#buildblocks
        self.blocks = nn.ModuleList([SwinTransformerBlock(
                dim=dim,
                num_heads=num_heads,
                window_size=window_size,
                shift_size=0if(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]ifisinstance(drop_path, list)else drop_path,
                norm_layer=norm_layer,
                Global=False)for i in range(depth)])#patchmerging layerif 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):#calculateattention 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 =0for 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#assertL == 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#ifself.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
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):returndrop_path_f(x, self.drop_prob, self.training)
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

四、引入解耦头部层

引入解耦头部的方法，就不在讲述了，其他比我厉害的博主也都有些写引入解耦头部的方法，大家可以去网上查看别的博客。

五、修改模型yaml文件

这里是我修改的模型yaml文件如下： 自己添加的注意力机制层数稍微多点。从下面可以看出在在head部分每一层的SwinTransformer_Layer都加了一层注意力机制。最后一层Detect部分采用Decoupled_Detect解耦头部。

#YOLOv5 🚀 by Ultralytics, GPL-3.0 license#Parameters
nc:1 # 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 v6.0 backbone
backbone:
  # [from, number, module, args][[-1,1, Conv,[64,6,2,2]],                            # 0-P1/2[-1,1, Conv,[128,3,2]],                              # 1-P2/4[-1,3, C3,[128]],                                      # 2[-1,1, Conv,[256,3,2]],                              # 3-P3/8[-1,1, SwinTransformer_Layer,[128,2,8,True,8]],        # 4[-1,1, Conv,[512,3,2]],                              # 5-P4/16[-1,1, SwinTransformer_Layer,[256,2,8,True,8]],        # 6[-1,1, Conv,[1024,3,2]],                             # 7-P5/32[-1,1, SwinTransformer_Layer,[512,2,8,True,4]],        # 8[-1,1, GAM_Attention,[512,512]],                       # 9[-1,1, SPPF,[1024,5]],                                # 10]
head:[[-1,1, Conv,[512,1,1]],                              # 11[-1,1, nn.Upsample,[None,2,'nearest']],              # 12[[-1,6],1, Concat,[1]],                               # 13[-1,3, C3,[512, False]],                               # 14[-1,1, Conv,[256,1,1]],                              # 15[-1,1, nn.Upsample,[None,2,'nearest']],              # 16[[-1,4],1, Concat,[1]],                               # 17[-1,3, C3,[256, False]],                               # 18[-1,1,GAM_Attention,[128,128]],                        # 19[-1,1, Conv,[512,3,2]],                              # 20[[-1,6,13],1, Concat,[1]],                           # 21[-1,3, C3,[512, False]],                               # 22[-1,1, SwinTransformer_Layer,[256,2,2,True,8]],        # 23[-1,1,GAM_Attention,[256,256]],                        # 24[-1,1, Conv,[1024,3,2]],                             # 25[[-1,10],1, Concat,[1]],                              # 26[-1,3, C3,[1024, False]],                              # 27(P4/16-medium)[-1,1, SwinTransformer_Layer,[512,2,2,True,4]],        # 28[-1,1,GAM_Attention,[512,512]],                        # 29[[19,24,29],1,Decoupled_Detect,[nc, anchors]],       # Detect(P3, P4, P5)]

六、运行代码

1.train.py报错问题

在修改后的YOLO v5中运行上述修改的yaml模型文件时候训练会报错，因为精度的问题，报错内容如下：RuntimeError: expected scalar type Half but found Float
在train.py的def parse_opt(known=False)中增加下面的最后一行代码：

添加上述代码还会继续报错，还是同样的问题，在对train.py文件进行继续修改，在下面两个部分添加half=not opt.swin_float这句代码修改后部分如下：

2.再次运行train.py

再次运行代码，–data部分是我自己的数据集yam文件，–cfg部分是上述修改后的模型yaml文件，这里权重设置为空，训练为200次，批次大小为4。终端运行训练代码如下：

python train.py  --data  huo.yaml  --cfg yolov5s_swin_head.yaml --weights ' '--epoch 200--batch-size 4--swin_float

3.自己数据集实验结果

模型文件mAPmAP@.5:.95YOLO v575.8%45.9%YOLO swin-head76.1%47.9%

总结

第一次写这个，奈何自己才疏学浅，后期有新的想法会继续更新。还望大家积极批评指正。
commom.py部分代码参考：
链接: [link] (https://github.com/iloveai8086/YOLOC)