2024.4.28
兄弟们注意一下：因为ultralytics使用的一键安装模式，本地再下载，会和环境里的包冲突。

我的方法是将ultralytics替换环境里的ultralytics包，然后更改ultralytics包里的文件、代码

做法：

比如我使用的Ubuntu，于是只要将我新上传的ultralytics文件夹替换掉\wsl.localhost\Ubuntu-20.04\home\ling\miniconda3\envs\torch\lib\python3.8\site-packages目录下的ultralytics即可

一、改进的操作方法

1.注册模块（以SPPFCSPC为例）

在nn/modules/block.py里注册要使用的新模块
有的在block里，有的在conv里，比如深度卷积等等需要在conv里

插入下面的代码：

2.各种引用都加上该模块

1. 翻到block.py最上面，加上模块名
   
2. 打开nn/modules/init.py
   需要注意的是，加入两个地方：
   ①之前模块注册在哪个py文件，就加入到哪里
   ②__all__的地方添加
   
3. 打开nn/tasks.py
   
   在ultralytics.nn.modules里import新添加的模块
   
   之后翻到下面649行左右，像图中这样添加
   
   注意： 简单的模块这样添加就注册完成了，但是如果是注意力机制，有的需要在nn/tasks.py再添加一些东西，比如下面这两个注意力机制ECAAttention 和 ShuffleAttention
   

3.在yaml文件里使用

打开cfg/models/v8，创建一个新的yaml
在你选择好的s、m或x的基础上进行修改，如下：

二、注意事项

1. 固定的模块只能由某些模块替换，不能随意替换

如：SPPF应替换为SPPCSPC、SPPCSPC_group、SPP等
格式如下：

  - [-1, 1, SPPCSPC_group, [1024]]  # 11
  
  - [-1, 1, SPPCSPC, [1024]]  # 11

如：Conv由DWConv、GhostConv等conv替换，conv.py里有很多模块

因为有些模块有特定的格式，随意替换，不改其他，容易报错，最好还是看我的yaml文件里是如何改的

2. DWConv、GhostConv可以减小模型复杂度，提高速度，精度可能会下降
3. 模块可以自由替换几个，可以替换所有可以替换的，也可以只替换某几个
   比如下面这样

## Ultralytics YOLO 🚀, GPL-3.0 license
# Parameters
nc: 1  # number of classes
depth_multiple: 1.00  # scales module repeats
width_multiple: 1.25  # scales convolution channels

# YOLOv8.0x6 backbone
backbone:
  # [from, repeats, module, args]
  - [-1, 1, Focus, [64, 3]]  # 0-P1/2
  - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
  - [-1, 3, C2f, [128, True]]
  - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
  - [-1, 6, C2f, [256, True]]
  - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
  - [-1, 6, C2f, [512, True]]
  - [-1, 1, Conv, [768, 3, 2]]  # 7-P5/32
  - [-1, 3, C2f, [768, True]]
  - [-1, 1, Conv, [1024, 3, 2]]  # 9-P6/64
  - [-1, 3, C2f, [1024, True]]
  - [-1, 1, SPPCSPC, [1024]]  # 11

# YOLOv8.0x6 head
head:
  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
  - [[-1, 8], 1, Concat, [1]]  # cat backbone P5
  - [-1, 3, C2, [768, False]]  # 14

  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
  - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
  - [-1, 3, C2, [512, False]]  # 17

  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
  - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
  - [-1, 3, C2, [256, False]]  # 20 (P3/8-small)

  - [-1, 1, GhostConv, [256, 3, 2]]
  - [[-1, 17], 1, Concat, [1]]  # cat head P4
  - [-1, 3, C2, [512, False]]  # 23 (P4/16-medium)

  - [-1, 1, GhostConv, [512, 3, 2]]
  - [[-1, 14], 1, Concat, [1]]  # cat head P5
  - [-1, 3, C2, [768, False]]  # 26 (P5/32-large)

  - [-1, 1, GhostConv, [768, 3, 2]]
  - [[-1, 11], 1, Concat, [1]]  # cat head P6
  - [-1, 3, C2, [1024, False]]  # 29 (P6/64-xlarge)

  - [[20, 23, 26, 29], 1, Detect, [nc]]  # Detect(P3, P4, P5, P6)

或者这个，可以替换两个

可以替换三个

三、众多模块分享

swin transformer模块

#----------swintf-----------------------------------------
class SwinTransformerBlock(nn.Module):
    def __init__(self, c1, c2, num_heads, num_layers, window_size=8):
        super().__init__()
        self.conv = None
        if c1 != c2:
            self.conv = Conv(c1, c2)

        # remove input_resolution
        self.blocks = nn.Sequential(*[SwinTransformerLayer(dim=c2, num_heads=num_heads, window_size=window_size,
                                                           shift_size=0 if (i % 2 == 0) else window_size // 2) for i in
                                      range(num_layers)])

    def forward(self, x):
        if self.conv is not None:
            x = self.conv(x)
        x = self.blocks(x)
        return x


class WindowAttention(nn.Module):

    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

        # define a parameter table of relative position bias
        self.relative_position_bias_table = nn.Parameter(
            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads))  # 2*Wh-1 * 2*Ww-1, nH

        # get pair-wise relative position index for each token inside the window
        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, 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.window_size[0] - 1  # shift to start from 0
        relative_coords[:, :, 1] += self.window_size[1] - 1
        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
        self.register_buffer("relative_position_index", relative_position_index)

        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.normal_(self.relative_position_bias_table, std=.02)
        self.softmax = nn.Softmax(dim=-1)

    def forward(self, x, mask=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]  # make torchscript happy (cannot use tensor as tuple)

        q = q * self.scale
        attn = (q @ k.transpose(-2, -1))

        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
        attn = attn + relative_position_bias.unsqueeze(0)

        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)

        # print(attn.dtype, v.dtype)
        try:
            x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        except:
            # print(attn.dtype, v.dtype)
            x = (attn.half() @ v).transpose(1, 2).reshape(B_, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


class Mlp(nn.Module):

    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.SiLU, 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


class SwinTransformerLayer(nn.Module):

    def __init__(self, dim, num_heads, window_size=8, shift_size=0,
                 mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0.,
                 act_layer=nn.SiLU, norm_layer=nn.LayerNorm):
        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
        # if min(self.input_resolution) <= self.window_size:
        #     # if window size is larger than input resolution, we don't partition windows
        #     self.shift_size = 0
        #     self.window_size = min(self.input_resolution)
        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"

        self.norm1 = norm_layer(dim)
        self.attn = WindowAttention(
            dim, window_size=(self.window_size, 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)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

    def create_mask(self, H, W):
        # calculate attention mask for SW-MSA
        img_mask = torch.zeros((1, H, W, 1))  # 1 H W 1
        h_slices = (slice(0, -self.window_size),
                    slice(-self.window_size, -self.shift_size),
                    slice(-self.shift_size, None))
        w_slices = (slice(0, -self.window_size),
                    slice(-self.window_size, -self.shift_size),
                    slice(-self.shift_size, None))
        cnt = 0
        for h in h_slices:
            for w in w_slices:
                img_mask[:, h, w, :] = cnt
                cnt += 1

        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

        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

        mask_windows = window_partition(img_mask, self.window_size)  # nW, window_size, window_size, 1
        mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
        attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
        attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0))

        return attn_mask

    def forward(self, x):
        # reshape x[b c h w] to x[b l c]
        _, _, H_, W_ = x.shape

        Padding = False
        if min(H_, W_) < self.window_size or H_ % self.window_size != 0 or W_ % self.window_size != 0:
            Padding = True
            # print(f'img_size {min(H_, W_)} is less than (or not divided by) window_size {self.window_size}, Padding.')
            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, pad_r, 0, pad_b))

        # print('2', x.shape)
        B, C, H, W = x.shape
        L = H * W
        x = x.permute(0, 2, 3, 1).contiguous().view(B, L, C)  # b, L, c

        # create mask from init to forward
        if self.shift_size > 0:
            attn_mask = self.create_mask(H, W).to(x.device)
        else:
            attn_mask = None

        shortcut = x
        x = self.norm1(x)
        x = x.view(B, H, W, C)

        # 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

        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

        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

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

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

        # merge windows
        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
        shifted_x = window_reverse(attn_windows, self.window_size, H, W)  # B H' W' C

        # reverse cyclic shift
        if self.shift_size > 0:
            x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
        else:
            x = shifted_x
        x = x.view(B, H * W, C)

        # FFN
        x = shortcut + self.drop_path(x)
        x = x + self.drop_path(self.mlp(self.norm2(x)))

        x = x.permute(0, 2, 1).contiguous().view(-1, C, H, W)  # b c h w

        if Padding:
            x = x[:, :, :H_, :W_]  # reverse padding

        return x


class C3STR(C3):
    # C3 module with SwinTransformerBlock()
    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        super().__init__(c1, c2, n, shortcut, g, e)
        c_ = int(c2 * e)
        num_heads = c_ // 32
        self.m = SwinTransformerBlock(c_, c_, num_heads, n)

名称叫C3STR
下面为我使用的比较好的搭配：

## Ultralytics YOLO , GPL-3.0 license
# Parameters
nc: 1  # number of classes
depth_multiple: 1.00  # scales module repeats
width_multiple: 1.25  # scales convolution channels

# YOLOv8.0x6 backbone
backbone:
  # [from, repeats, module, args]
  - [-1, 1, Focus, [64, 3]]  # 0-P1/2
  - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
  - [-1, 3, C3STR, [128]]
  - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
  - [-1, 6, C3STR, [256]]
  - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
  - [-1, 6, C3STR, [512]]
  - [-1, 1, Conv, [768, 3, 2]]  # 7-P5/32
  - [-1, 3, C3STR, [768]]
  - [-1, 1, Conv, [1024, 3, 2]]  # 9-P6/64
  - [-1, 3, C2f, [1024, True]]
  - [-1, 1, SPPCSPC, [1024, 5]]  # 11

# YOLOv8.0x6 head
head:
  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
  - [[-1, 8], 1, Concat, [1]]  # cat backbone P5
  - [-1, 3, C2, [768, False]]  # 14

  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
  - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
  - [-1, 3, C2, [512, False]]  # 17

  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
  - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
  - [-1, 3, C2, [256, False]]  # 20 (P3/8-small)

  - [-1, 1, DWConv, [256, 3, 2]]
  - [[-1, 17], 1, Concat, [1]]  # cat head P4
  - [-1, 3, C2, [512, False]]  # 23 (P4/16-medium)

  - [-1, 1, DWConv, [512, 3, 2]]
  - [[-1, 14], 1, Concat, [1]]  # cat head P5
  - [-1, 3, C2, [768, False]]  # 26 (P5/32-large)

  - [-1, 1, DWConv, [768, 3, 2]]
  - [[-1, 11], 1, Concat, [1]]  # cat head P6
  - [-1, 3, C2, [1024, False]]  # 29 (P6/64-xlarge)

  - [[20, 23, 26, 29], 1, Detect, [nc]]  # Detect(P3, P4, P5, P6)

注：swin transformer确实会提高一些精度，但模型复杂度会提高，并且需要耗费资源，最好3090之类的才能跑C3STR模块多的网络结构。

我有众多的搭配方案在我的GitHub里
https://github.com/ssrzero123/STF-YOLO
如果对大家有用的话希望得到一个星星star收藏（拜托啦）

如下是我训练的效果，绝对真实，诸位可以浅浅参考一下
（训练的策略好像大部分都是改成了wiou，不过影响不大）