跳至内容

参考资料 ultralytics/nn/modules/block.py

备注

该文件可从https://github.com/ultralytics/ultralytics/blob/main/ ultralytics/nn/modules/block .py 获取。如果您发现问题,请通过提交 Pull Request🛠️ 帮助修复。谢谢🙏!



ultralytics.nn.modules.block.DFL

垒球 Module

分布焦距损耗(DFL)积分模块。

建议在全身病灶丧失 https://ieeexplore.ieee.org/document/9792391

源代码 ultralytics/nn/modules/block.py
class DFL(nn.Module):
    """
    Integral module of Distribution Focal Loss (DFL).

    Proposed in Generalized Focal Loss https://ieeexplore.ieee.org/document/9792391
    """

    def __init__(self, c1=16):
        """Initialize a convolutional layer with a given number of input channels."""
        super().__init__()
        self.conv = nn.Conv2d(c1, 1, 1, bias=False).requires_grad_(False)
        x = torch.arange(c1, dtype=torch.float)
        self.conv.weight.data[:] = nn.Parameter(x.view(1, c1, 1, 1))
        self.c1 = c1

    def forward(self, x):
        """Applies a transformer layer on input tensor 'x' and returns a tensor."""
        b, _, a = x.shape  # batch, channels, anchors
        return self.conv(x.view(b, 4, self.c1, a).transpose(2, 1).softmax(1)).view(b, 4, a)

__init__(c1=16)

用给定的输入通道数初始化卷积层。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1=16):
    """Initialize a convolutional layer with a given number of input channels."""
    super().__init__()
    self.conv = nn.Conv2d(c1, 1, 1, bias=False).requires_grad_(False)
    x = torch.arange(c1, dtype=torch.float)
    self.conv.weight.data[:] = nn.Parameter(x.view(1, c1, 1, 1))
    self.c1 = c1

forward(x)

在输入tensor 'x' 上应用转换层,并返回tensor 。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Applies a transformer layer on input tensor 'x' and returns a tensor."""
    b, _, a = x.shape  # batch, channels, anchors
    return self.conv(x.view(b, 4, self.c1, a).transpose(2, 1).softmax(1)).view(b, 4, a)



ultralytics.nn.modules.block.Proto

垒球 Module

YOLOv8 用于分割模型的掩码 Proto 模块。

源代码 ultralytics/nn/modules/block.py
class Proto(nn.Module):
    """YOLOv8 mask Proto module for segmentation models."""

    def __init__(self, c1, c_=256, c2=32):
        """
        Initializes the YOLOv8 mask Proto module with specified number of protos and masks.

        Input arguments are ch_in, number of protos, number of masks.
        """
        super().__init__()
        self.cv1 = Conv(c1, c_, k=3)
        self.upsample = nn.ConvTranspose2d(c_, c_, 2, 2, 0, bias=True)  # nn.Upsample(scale_factor=2, mode='nearest')
        self.cv2 = Conv(c_, c_, k=3)
        self.cv3 = Conv(c_, c2)

    def forward(self, x):
        """Performs a forward pass through layers using an upsampled input image."""
        return self.cv3(self.cv2(self.upsample(self.cv1(x))))

__init__(c1, c_=256, c2=32)

使用指定数量的原语和掩码初始化YOLOv8 mask Proto 模块。

输入参数包括 ch_in、protos 数量、掩码数量。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c_=256, c2=32):
    """
    Initializes the YOLOv8 mask Proto module with specified number of protos and masks.

    Input arguments are ch_in, number of protos, number of masks.
    """
    super().__init__()
    self.cv1 = Conv(c1, c_, k=3)
    self.upsample = nn.ConvTranspose2d(c_, c_, 2, 2, 0, bias=True)  # nn.Upsample(scale_factor=2, mode='nearest')
    self.cv2 = Conv(c_, c_, k=3)
    self.cv3 = Conv(c_, c2)

forward(x)

使用上采样输入图像执行层前传。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Performs a forward pass through layers using an upsampled input image."""
    return self.cv3(self.cv2(self.upsample(self.cv1(x))))



ultralytics.nn.modules.block.HGStem

垒球 Module

带有 5 个卷积和一个 maxpool2d 的 PPHGNetV2 的 StemBlock。

https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/backbones/hgnet_v2.py

源代码 ultralytics/nn/modules/block.py
class HGStem(nn.Module):
    """
    StemBlock of PPHGNetV2 with 5 convolutions and one maxpool2d.

    https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/backbones/hgnet_v2.py
    """

    def __init__(self, c1, cm, c2):
        """Initialize the SPP layer with input/output channels and specified kernel sizes for max pooling."""
        super().__init__()
        self.stem1 = Conv(c1, cm, 3, 2, act=nn.ReLU())
        self.stem2a = Conv(cm, cm // 2, 2, 1, 0, act=nn.ReLU())
        self.stem2b = Conv(cm // 2, cm, 2, 1, 0, act=nn.ReLU())
        self.stem3 = Conv(cm * 2, cm, 3, 2, act=nn.ReLU())
        self.stem4 = Conv(cm, c2, 1, 1, act=nn.ReLU())
        self.pool = nn.MaxPool2d(kernel_size=2, stride=1, padding=0, ceil_mode=True)

    def forward(self, x):
        """Forward pass of a PPHGNetV2 backbone layer."""
        x = self.stem1(x)
        x = F.pad(x, [0, 1, 0, 1])
        x2 = self.stem2a(x)
        x2 = F.pad(x2, [0, 1, 0, 1])
        x2 = self.stem2b(x2)
        x1 = self.pool(x)
        x = torch.cat([x1, x2], dim=1)
        x = self.stem3(x)
        x = self.stem4(x)
        return x

__init__(c1, cm, c2)

使用输入/输出通道和指定的内核大小初始化 SPP 层,以实现最大池化。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, cm, c2):
    """Initialize the SPP layer with input/output channels and specified kernel sizes for max pooling."""
    super().__init__()
    self.stem1 = Conv(c1, cm, 3, 2, act=nn.ReLU())
    self.stem2a = Conv(cm, cm // 2, 2, 1, 0, act=nn.ReLU())
    self.stem2b = Conv(cm // 2, cm, 2, 1, 0, act=nn.ReLU())
    self.stem3 = Conv(cm * 2, cm, 3, 2, act=nn.ReLU())
    self.stem4 = Conv(cm, c2, 1, 1, act=nn.ReLU())
    self.pool = nn.MaxPool2d(kernel_size=2, stride=1, padding=0, ceil_mode=True)

forward(x)

PPHGNetV2 主干层的前向通道。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass of a PPHGNetV2 backbone layer."""
    x = self.stem1(x)
    x = F.pad(x, [0, 1, 0, 1])
    x2 = self.stem2a(x)
    x2 = F.pad(x2, [0, 1, 0, 1])
    x2 = self.stem2b(x2)
    x1 = self.pool(x)
    x = torch.cat([x1, x2], dim=1)
    x = self.stem3(x)
    x = self.stem4(x)
    return x



ultralytics.nn.modules.block.HGBlock

垒球 Module

带有 2 个卷积和 LightConv 的 PPHGNetV2 的 HG_Block。

https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/backbones/hgnet_v2.py

源代码 ultralytics/nn/modules/block.py
class HGBlock(nn.Module):
    """
    HG_Block of PPHGNetV2 with 2 convolutions and LightConv.

    https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/backbones/hgnet_v2.py
    """

    def __init__(self, c1, cm, c2, k=3, n=6, lightconv=False, shortcut=False, act=nn.ReLU()):
        """Initializes a CSP Bottleneck with 1 convolution using specified input and output channels."""
        super().__init__()
        block = LightConv if lightconv else Conv
        self.m = nn.ModuleList(block(c1 if i == 0 else cm, cm, k=k, act=act) for i in range(n))
        self.sc = Conv(c1 + n * cm, c2 // 2, 1, 1, act=act)  # squeeze conv
        self.ec = Conv(c2 // 2, c2, 1, 1, act=act)  # excitation conv
        self.add = shortcut and c1 == c2

    def forward(self, x):
        """Forward pass of a PPHGNetV2 backbone layer."""
        y = [x]
        y.extend(m(y[-1]) for m in self.m)
        y = self.ec(self.sc(torch.cat(y, 1)))
        return y + x if self.add else y

__init__(c1, cm, c2, k=3, n=6, lightconv=False, shortcut=False, act=nn.ReLU())

使用指定的输入和输出通道,以 1 次卷积初始化 CSP 瓶颈。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, cm, c2, k=3, n=6, lightconv=False, shortcut=False, act=nn.ReLU()):
    """Initializes a CSP Bottleneck with 1 convolution using specified input and output channels."""
    super().__init__()
    block = LightConv if lightconv else Conv
    self.m = nn.ModuleList(block(c1 if i == 0 else cm, cm, k=k, act=act) for i in range(n))
    self.sc = Conv(c1 + n * cm, c2 // 2, 1, 1, act=act)  # squeeze conv
    self.ec = Conv(c2 // 2, c2, 1, 1, act=act)  # excitation conv
    self.add = shortcut and c1 == c2

forward(x)

PPHGNetV2 主干层的前向通道。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass of a PPHGNetV2 backbone layer."""
    y = [x]
    y.extend(m(y[-1]) for m in self.m)
    y = self.ec(self.sc(torch.cat(y, 1)))
    return y + x if self.add else y



ultralytics.nn.modules.block.SPP

垒球 Module

空间金字塔汇集(SPP)层 https://arxiv.org/abs/1406.4729。

源代码 ultralytics/nn/modules/block.py
class SPP(nn.Module):
    """Spatial Pyramid Pooling (SPP) layer https://arxiv.org/abs/1406.4729."""

    def __init__(self, c1, c2, k=(5, 9, 13)):
        """Initialize the SPP layer with input/output channels and pooling kernel sizes."""
        super().__init__()
        c_ = c1 // 2  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1)
        self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k])

    def forward(self, x):
        """Forward pass of the SPP layer, performing spatial pyramid pooling."""
        x = self.cv1(x)
        return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1))

__init__(c1, c2, k=(5, 9, 13))

用输入/输出通道和池核大小初始化 SPP 层。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, k=(5, 9, 13)):
    """Initialize the SPP layer with input/output channels and pooling kernel sizes."""
    super().__init__()
    c_ = c1 // 2  # hidden channels
    self.cv1 = Conv(c1, c_, 1, 1)
    self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1)
    self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k])

forward(x)

SPP 层的前向传递,执行空间金字塔汇集。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass of the SPP layer, performing spatial pyramid pooling."""
    x = self.cv1(x)
    return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1))



ultralytics.nn.modules.block.SPPF

垒球 Module

空间金字塔池化--快速(SPPF)层,YOLOv5 ,作者 Glenn Jocher。

源代码 ultralytics/nn/modules/block.py
class SPPF(nn.Module):
    """Spatial Pyramid Pooling - Fast (SPPF) layer for YOLOv5 by Glenn Jocher."""

    def __init__(self, c1, c2, k=5):
        """
        Initializes the SPPF layer with given input/output channels and kernel size.

        This module is equivalent to SPP(k=(5, 9, 13)).
        """
        super().__init__()
        c_ = c1 // 2  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c_ * 4, c2, 1, 1)
        self.m = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)

    def forward(self, x):
        """Forward pass through Ghost Convolution block."""
        y = [self.cv1(x)]
        y.extend(self.m(y[-1]) for _ in range(3))
        return self.cv2(torch.cat(y, 1))

__init__(c1, c2, k=5)

使用给定的输入/输出通道和内核大小初始化 SPPF 层。

该模块等同于 SPP(k=(5, 9, 13))。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, k=5):
    """
    Initializes the SPPF layer with given input/output channels and kernel size.

    This module is equivalent to SPP(k=(5, 9, 13)).
    """
    super().__init__()
    c_ = c1 // 2  # hidden channels
    self.cv1 = Conv(c1, c_, 1, 1)
    self.cv2 = Conv(c_ * 4, c2, 1, 1)
    self.m = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)

forward(x)

前向通过幽灵卷积块。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through Ghost Convolution block."""
    y = [self.cv1(x)]
    y.extend(self.m(y[-1]) for _ in range(3))
    return self.cv2(torch.cat(y, 1))



ultralytics.nn.modules.block.C1

垒球 Module

1 次卷积的 CSP 瓶颈。

源代码 ultralytics/nn/modules/block.py
class C1(nn.Module):
    """CSP Bottleneck with 1 convolution."""

    def __init__(self, c1, c2, n=1):
        """Initializes the CSP Bottleneck with configurations for 1 convolution with arguments ch_in, ch_out, number."""
        super().__init__()
        self.cv1 = Conv(c1, c2, 1, 1)
        self.m = nn.Sequential(*(Conv(c2, c2, 3) for _ in range(n)))

    def forward(self, x):
        """Applies cross-convolutions to input in the C3 module."""
        y = self.cv1(x)
        return self.m(y) + y

__init__(c1, c2, n=1)

使用参数 ch_in、ch_out、number 初始化 CSP Bottleneck,配置 1 次卷积。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1):
    """Initializes the CSP Bottleneck with configurations for 1 convolution with arguments ch_in, ch_out, number."""
    super().__init__()
    self.cv1 = Conv(c1, c2, 1, 1)
    self.m = nn.Sequential(*(Conv(c2, c2, 3) for _ in range(n)))

forward(x)

对 C3 模块中的输入进行交叉旋转。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Applies cross-convolutions to input in the C3 module."""
    y = self.cv1(x)
    return self.m(y) + y



ultralytics.nn.modules.block.C2

垒球 Module

有 2 个卷积的 CSP 瓶颈。

源代码 ultralytics/nn/modules/block.py
class C2(nn.Module):
    """CSP Bottleneck with 2 convolutions."""

    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        """Initializes the CSP Bottleneck with 2 convolutions module with arguments ch_in, ch_out, number, shortcut,
        groups, expansion.
        """
        super().__init__()
        self.c = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, 2 * self.c, 1, 1)
        self.cv2 = Conv(2 * self.c, c2, 1)  # optional act=FReLU(c2)
        # self.attention = ChannelAttention(2 * self.c)  # or SpatialAttention()
        self.m = nn.Sequential(*(Bottleneck(self.c, self.c, shortcut, g, k=((3, 3), (3, 3)), e=1.0) for _ in range(n)))

    def forward(self, x):
        """Forward pass through the CSP bottleneck with 2 convolutions."""
        a, b = self.cv1(x).chunk(2, 1)
        return self.cv2(torch.cat((self.m(a), b), 1))

__init__(c1, c2, n=1, shortcut=True, g=1, e=0.5)

使用参数 ch_in、ch_out、number、快捷方式初始化 CSP Bottleneck 的 2 个卷积模块、 组、扩展。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
    """Initializes the CSP Bottleneck with 2 convolutions module with arguments ch_in, ch_out, number, shortcut,
    groups, expansion.
    """
    super().__init__()
    self.c = int(c2 * e)  # hidden channels
    self.cv1 = Conv(c1, 2 * self.c, 1, 1)
    self.cv2 = Conv(2 * self.c, c2, 1)  # optional act=FReLU(c2)
    # self.attention = ChannelAttention(2 * self.c)  # or SpatialAttention()
    self.m = nn.Sequential(*(Bottleneck(self.c, self.c, shortcut, g, k=((3, 3), (3, 3)), e=1.0) for _ in range(n)))

forward(x)

前向通过 CSP 瓶颈,有 2 个卷积。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through the CSP bottleneck with 2 convolutions."""
    a, b = self.cv1(x).chunk(2, 1)
    return self.cv2(torch.cat((self.m(a), b), 1))



ultralytics.nn.modules.block.C2f

垒球 Module

用 2 次卷积更快地实现 CSP 瓶颈。

源代码 ultralytics/nn/modules/block.py
class C2f(nn.Module):
    """Faster Implementation of CSP Bottleneck with 2 convolutions."""

    def __init__(self, c1, c2, n=1, shortcut=False, g=1, e=0.5):
        """Initialize CSP bottleneck layer with two convolutions with arguments ch_in, ch_out, number, shortcut, groups,
        expansion.
        """
        super().__init__()
        self.c = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, 2 * self.c, 1, 1)
        self.cv2 = Conv((2 + n) * self.c, c2, 1)  # optional act=FReLU(c2)
        self.m = nn.ModuleList(Bottleneck(self.c, self.c, shortcut, g, k=((3, 3), (3, 3)), e=1.0) for _ in range(n))

    def forward(self, x):
        """Forward pass through C2f layer."""
        y = list(self.cv1(x).chunk(2, 1))
        y.extend(m(y[-1]) for m in self.m)
        return self.cv2(torch.cat(y, 1))

    def forward_split(self, x):
        """Forward pass using split() instead of chunk()."""
        y = list(self.cv1(x).split((self.c, self.c), 1))
        y.extend(m(y[-1]) for m in self.m)
        return self.cv2(torch.cat(y, 1))

__init__(c1, c2, n=1, shortcut=False, g=1, e=0.5)

用两个卷积初始化 CSP 瓶颈层,参数为 ch_in、ch_out、number、shortcut、group、 扩展。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, shortcut=False, g=1, e=0.5):
    """Initialize CSP bottleneck layer with two convolutions with arguments ch_in, ch_out, number, shortcut, groups,
    expansion.
    """
    super().__init__()
    self.c = int(c2 * e)  # hidden channels
    self.cv1 = Conv(c1, 2 * self.c, 1, 1)
    self.cv2 = Conv((2 + n) * self.c, c2, 1)  # optional act=FReLU(c2)
    self.m = nn.ModuleList(Bottleneck(self.c, self.c, shortcut, g, k=((3, 3), (3, 3)), e=1.0) for _ in range(n))

forward(x)

向前穿过 C2f 层。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through C2f layer."""
    y = list(self.cv1(x).chunk(2, 1))
    y.extend(m(y[-1]) for m in self.m)
    return self.cv2(torch.cat(y, 1))

forward_split(x)

使用 split() 代替 chunk() 向前传递。

源代码 ultralytics/nn/modules/block.py
def forward_split(self, x):
    """Forward pass using split() instead of chunk()."""
    y = list(self.cv1(x).split((self.c, self.c), 1))
    y.extend(m(y[-1]) for m in self.m)
    return self.cv2(torch.cat(y, 1))



ultralytics.nn.modules.block.C3

垒球 Module

有 3 个卷积的 CSP 瓶颈。

源代码 ultralytics/nn/modules/block.py
class C3(nn.Module):
    """CSP Bottleneck with 3 convolutions."""

    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        """Initialize the CSP Bottleneck with given channels, number, shortcut, groups, and expansion values."""
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = Conv(c1, c_, 1, 1)
        self.cv3 = Conv(2 * c_, c2, 1)  # optional act=FReLU(c2)
        self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, k=((1, 1), (3, 3)), e=1.0) for _ in range(n)))

    def forward(self, x):
        """Forward pass through the CSP bottleneck with 2 convolutions."""
        return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), 1))

__init__(c1, c2, n=1, shortcut=True, g=1, e=0.5)

使用给定的通道、编号、快捷方式、分组和扩展值初始化 CSP Bottleneck。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
    """Initialize the CSP Bottleneck with given channels, number, shortcut, groups, and expansion values."""
    super().__init__()
    c_ = int(c2 * e)  # hidden channels
    self.cv1 = Conv(c1, c_, 1, 1)
    self.cv2 = Conv(c1, c_, 1, 1)
    self.cv3 = Conv(2 * c_, c2, 1)  # optional act=FReLU(c2)
    self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, k=((1, 1), (3, 3)), e=1.0) for _ in range(n)))

forward(x)

前向通过 CSP 瓶颈,有 2 个卷积。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through the CSP bottleneck with 2 convolutions."""
    return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), 1))



ultralytics.nn.modules.block.C3x

垒球 C3

C3 模块带交叉旋转。

源代码 ultralytics/nn/modules/block.py
class C3x(C3):
    """C3 module with cross-convolutions."""

    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        """Initialize C3TR instance and set default parameters."""
        super().__init__(c1, c2, n, shortcut, g, e)
        self.c_ = int(c2 * e)
        self.m = nn.Sequential(*(Bottleneck(self.c_, self.c_, shortcut, g, k=((1, 3), (3, 1)), e=1) for _ in range(n)))

__init__(c1, c2, n=1, shortcut=True, g=1, e=0.5)

初始化 C3TR 实例并设置默认参数。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
    """Initialize C3TR instance and set default parameters."""
    super().__init__(c1, c2, n, shortcut, g, e)
    self.c_ = int(c2 * e)
    self.m = nn.Sequential(*(Bottleneck(self.c_, self.c_, shortcut, g, k=((1, 3), (3, 1)), e=1) for _ in range(n)))



ultralytics.nn.modules.block.RepC3

垒球 Module

代表 C3。

源代码 ultralytics/nn/modules/block.py
class RepC3(nn.Module):
    """Rep C3."""

    def __init__(self, c1, c2, n=3, e=1.0):
        """Initialize CSP Bottleneck with a single convolution using input channels, output channels, and number."""
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, c2, 1, 1)
        self.cv2 = Conv(c1, c2, 1, 1)
        self.m = nn.Sequential(*[RepConv(c_, c_) for _ in range(n)])
        self.cv3 = Conv(c_, c2, 1, 1) if c_ != c2 else nn.Identity()

    def forward(self, x):
        """Forward pass of RT-DETR neck layer."""
        return self.cv3(self.m(self.cv1(x)) + self.cv2(x))

__init__(c1, c2, n=3, e=1.0)

使用输入通道、输出通道和编号对 CSP Bottleneck 进行一次卷积初始化。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=3, e=1.0):
    """Initialize CSP Bottleneck with a single convolution using input channels, output channels, and number."""
    super().__init__()
    c_ = int(c2 * e)  # hidden channels
    self.cv1 = Conv(c1, c2, 1, 1)
    self.cv2 = Conv(c1, c2, 1, 1)
    self.m = nn.Sequential(*[RepConv(c_, c_) for _ in range(n)])
    self.cv3 = Conv(c_, c2, 1, 1) if c_ != c2 else nn.Identity()

forward(x)

RT-DETR 。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass of RT-DETR neck layer."""
    return self.cv3(self.m(self.cv1(x)) + self.cv2(x))



ultralytics.nn.modules.block.C3TR

垒球 C3

使用 TransformerBlock() 的 C3 模块。

源代码 ultralytics/nn/modules/block.py
class C3TR(C3):
    """C3 module with TransformerBlock()."""

    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        """Initialize C3Ghost module with GhostBottleneck()."""
        super().__init__(c1, c2, n, shortcut, g, e)
        c_ = int(c2 * e)
        self.m = TransformerBlock(c_, c_, 4, n)

__init__(c1, c2, n=1, shortcut=True, g=1, e=0.5)

使用 GhostBottleneck() 初始化 C3Ghost 模块。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
    """Initialize C3Ghost module with GhostBottleneck()."""
    super().__init__(c1, c2, n, shortcut, g, e)
    c_ = int(c2 * e)
    self.m = TransformerBlock(c_, c_, 4, n)



ultralytics.nn.modules.block.C3Ghost

垒球 C3

使用 GhostBottleneck() 的 C3 模块。

源代码 ultralytics/nn/modules/block.py
class C3Ghost(C3):
    """C3 module with GhostBottleneck()."""

    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        """Initialize 'SPP' module with various pooling sizes for spatial pyramid pooling."""
        super().__init__(c1, c2, n, shortcut, g, e)
        c_ = int(c2 * e)  # hidden channels
        self.m = nn.Sequential(*(GhostBottleneck(c_, c_) for _ in range(n)))

__init__(c1, c2, n=1, shortcut=True, g=1, e=0.5)

初始化 "SPP "模块,为空间金字塔汇集法提供各种汇集规模。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
    """Initialize 'SPP' module with various pooling sizes for spatial pyramid pooling."""
    super().__init__(c1, c2, n, shortcut, g, e)
    c_ = int(c2 * e)  # hidden channels
    self.m = nn.Sequential(*(GhostBottleneck(c_, c_) for _ in range(n)))



ultralytics.nn.modules.block.GhostBottleneck

垒球 Module

幽灵瓶颈 https://github.com/huawei-noah/ghostnet。

源代码 ultralytics/nn/modules/block.py
class GhostBottleneck(nn.Module):
    """Ghost Bottleneck https://github.com/huawei-noah/ghostnet."""

    def __init__(self, c1, c2, k=3, s=1):
        """Initializes GhostBottleneck module with arguments ch_in, ch_out, kernel, stride."""
        super().__init__()
        c_ = c2 // 2
        self.conv = nn.Sequential(
            GhostConv(c1, c_, 1, 1),  # pw
            DWConv(c_, c_, k, s, act=False) if s == 2 else nn.Identity(),  # dw
            GhostConv(c_, c2, 1, 1, act=False),  # pw-linear
        )
        self.shortcut = (
            nn.Sequential(DWConv(c1, c1, k, s, act=False), Conv(c1, c2, 1, 1, act=False)) if s == 2 else nn.Identity()
        )

    def forward(self, x):
        """Applies skip connection and concatenation to input tensor."""
        return self.conv(x) + self.shortcut(x)

__init__(c1, c2, k=3, s=1)

使用参数 ch_in、ch_out、kernel 和 stride 初始化 GhostBottleneck 模块。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, k=3, s=1):
    """Initializes GhostBottleneck module with arguments ch_in, ch_out, kernel, stride."""
    super().__init__()
    c_ = c2 // 2
    self.conv = nn.Sequential(
        GhostConv(c1, c_, 1, 1),  # pw
        DWConv(c_, c_, k, s, act=False) if s == 2 else nn.Identity(),  # dw
        GhostConv(c_, c2, 1, 1, act=False),  # pw-linear
    )
    self.shortcut = (
        nn.Sequential(DWConv(c1, c1, k, s, act=False), Conv(c1, c2, 1, 1, act=False)) if s == 2 else nn.Identity()
    )

forward(x)

对输入tensor 应用跳转连接和连接。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Applies skip connection and concatenation to input tensor."""
    return self.conv(x) + self.shortcut(x)



ultralytics.nn.modules.block.Bottleneck

垒球 Module

标准瓶颈。

源代码 ultralytics/nn/modules/block.py
class Bottleneck(nn.Module):
    """Standard bottleneck."""

    def __init__(self, c1, c2, shortcut=True, g=1, k=(3, 3), e=0.5):
        """Initializes a bottleneck module with given input/output channels, shortcut option, group, kernels, and
        expansion.
        """
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, c_, k[0], 1)
        self.cv2 = Conv(c_, c2, k[1], 1, g=g)
        self.add = shortcut and c1 == c2

    def forward(self, x):
        """'forward()' applies the YOLO FPN to input data."""
        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))

__init__(c1, c2, shortcut=True, g=1, k=(3, 3), e=0.5)

使用给定的输入/输出通道、快捷选项、组、内核和扩展名初始化瓶颈模块。 扩展。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, shortcut=True, g=1, k=(3, 3), e=0.5):
    """Initializes a bottleneck module with given input/output channels, shortcut option, group, kernels, and
    expansion.
    """
    super().__init__()
    c_ = int(c2 * e)  # hidden channels
    self.cv1 = Conv(c1, c_, k[0], 1)
    self.cv2 = Conv(c_, c2, k[1], 1, g=g)
    self.add = shortcut and c1 == c2

forward(x)

forward() "将YOLO FPN 应用于输入数据。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """'forward()' applies the YOLO FPN to input data."""
    return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))



ultralytics.nn.modules.block.BottleneckCSP

垒球 Module

CSP 瓶颈 https://github.com/WongKinYiu/CrossStagePartialNetworks。

源代码 ultralytics/nn/modules/block.py
class BottleneckCSP(nn.Module):
    """CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks."""

    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        """Initializes the CSP Bottleneck given arguments for ch_in, ch_out, number, shortcut, groups, expansion."""
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, c_, 1, 1)
        self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False)
        self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False)
        self.cv4 = Conv(2 * c_, c2, 1, 1)
        self.bn = nn.BatchNorm2d(2 * c_)  # applied to cat(cv2, cv3)
        self.act = nn.SiLU()
        self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))

    def forward(self, x):
        """Applies a CSP bottleneck with 3 convolutions."""
        y1 = self.cv3(self.m(self.cv1(x)))
        y2 = self.cv2(x)
        return self.cv4(self.act(self.bn(torch.cat((y1, y2), 1))))

__init__(c1, c2, n=1, shortcut=True, g=1, e=0.5)

为 ch_in、ch_out、编号、快捷方式、分组和扩展参数初始化 CSP Bottleneck。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
    """Initializes the CSP Bottleneck given arguments for ch_in, ch_out, number, shortcut, groups, expansion."""
    super().__init__()
    c_ = int(c2 * e)  # hidden channels
    self.cv1 = Conv(c1, c_, 1, 1)
    self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False)
    self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False)
    self.cv4 = Conv(2 * c_, c2, 1, 1)
    self.bn = nn.BatchNorm2d(2 * c_)  # applied to cat(cv2, cv3)
    self.act = nn.SiLU()
    self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))

forward(x)

应用具有 3 个卷积的 CSP 瓶颈。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Applies a CSP bottleneck with 3 convolutions."""
    y1 = self.cv3(self.m(self.cv1(x)))
    y2 = self.cv2(x)
    return self.cv4(self.act(self.bn(torch.cat((y1, y2), 1))))



ultralytics.nn.modules.block.ResNetBlock

垒球 Module

具有标准卷积层的 ResNet 块。

源代码 ultralytics/nn/modules/block.py
class ResNetBlock(nn.Module):
    """ResNet block with standard convolution layers."""

    def __init__(self, c1, c2, s=1, e=4):
        """Initialize convolution with given parameters."""
        super().__init__()
        c3 = e * c2
        self.cv1 = Conv(c1, c2, k=1, s=1, act=True)
        self.cv2 = Conv(c2, c2, k=3, s=s, p=1, act=True)
        self.cv3 = Conv(c2, c3, k=1, act=False)
        self.shortcut = nn.Sequential(Conv(c1, c3, k=1, s=s, act=False)) if s != 1 or c1 != c3 else nn.Identity()

    def forward(self, x):
        """Forward pass through the ResNet block."""
        return F.relu(self.cv3(self.cv2(self.cv1(x))) + self.shortcut(x))

__init__(c1, c2, s=1, e=4)

使用给定参数初始化卷积。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, s=1, e=4):
    """Initialize convolution with given parameters."""
    super().__init__()
    c3 = e * c2
    self.cv1 = Conv(c1, c2, k=1, s=1, act=True)
    self.cv2 = Conv(c2, c2, k=3, s=s, p=1, act=True)
    self.cv3 = Conv(c2, c3, k=1, act=False)
    self.shortcut = nn.Sequential(Conv(c1, c3, k=1, s=s, act=False)) if s != 1 or c1 != c3 else nn.Identity()

forward(x)

前向通过 ResNet 区块。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through the ResNet block."""
    return F.relu(self.cv3(self.cv2(self.cv1(x))) + self.shortcut(x))



ultralytics.nn.modules.block.ResNetLayer

垒球 Module

具有多个 ResNet 块的 ResNet 层。

源代码 ultralytics/nn/modules/block.py
class ResNetLayer(nn.Module):
    """ResNet layer with multiple ResNet blocks."""

    def __init__(self, c1, c2, s=1, is_first=False, n=1, e=4):
        """Initializes the ResNetLayer given arguments."""
        super().__init__()
        self.is_first = is_first

        if self.is_first:
            self.layer = nn.Sequential(
                Conv(c1, c2, k=7, s=2, p=3, act=True), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
            )
        else:
            blocks = [ResNetBlock(c1, c2, s, e=e)]
            blocks.extend([ResNetBlock(e * c2, c2, 1, e=e) for _ in range(n - 1)])
            self.layer = nn.Sequential(*blocks)

    def forward(self, x):
        """Forward pass through the ResNet layer."""
        return self.layer(x)

__init__(c1, c2, s=1, is_first=False, n=1, e=4)

初始化给定参数的 ResNetLayer。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, s=1, is_first=False, n=1, e=4):
    """Initializes the ResNetLayer given arguments."""
    super().__init__()
    self.is_first = is_first

    if self.is_first:
        self.layer = nn.Sequential(
            Conv(c1, c2, k=7, s=2, p=3, act=True), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        )
    else:
        blocks = [ResNetBlock(c1, c2, s, e=e)]
        blocks.extend([ResNetBlock(e * c2, c2, 1, e=e) for _ in range(n - 1)])
        self.layer = nn.Sequential(*blocks)

forward(x)

前向通过 ResNet 层。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through the ResNet layer."""
    return self.layer(x)



ultralytics.nn.modules.block.MaxSigmoidAttnBlock

垒球 Module

最大西格蒙德注意块。

源代码 ultralytics/nn/modules/block.py
class MaxSigmoidAttnBlock(nn.Module):
    """Max Sigmoid attention block."""

    def __init__(self, c1, c2, nh=1, ec=128, gc=512, scale=False):
        """Initializes MaxSigmoidAttnBlock with specified arguments."""
        super().__init__()
        self.nh = nh
        self.hc = c2 // nh
        self.ec = Conv(c1, ec, k=1, act=False) if c1 != ec else None
        self.gl = nn.Linear(gc, ec)
        self.bias = nn.Parameter(torch.zeros(nh))
        self.proj_conv = Conv(c1, c2, k=3, s=1, act=False)
        self.scale = nn.Parameter(torch.ones(1, nh, 1, 1)) if scale else 1.0

    def forward(self, x, guide):
        """Forward process."""
        bs, _, h, w = x.shape

        guide = self.gl(guide)
        guide = guide.view(bs, -1, self.nh, self.hc)
        embed = self.ec(x) if self.ec is not None else x
        embed = embed.view(bs, self.nh, self.hc, h, w)

        aw = torch.einsum("bmchw,bnmc->bmhwn", embed, guide)
        aw = aw.max(dim=-1)[0]
        aw = aw / (self.hc**0.5)
        aw = aw + self.bias[None, :, None, None]
        aw = aw.sigmoid() * self.scale

        x = self.proj_conv(x)
        x = x.view(bs, self.nh, -1, h, w)
        x = x * aw.unsqueeze(2)
        return x.view(bs, -1, h, w)

__init__(c1, c2, nh=1, ec=128, gc=512, scale=False)

使用指定参数初始化 MaxSigmoidAttnBlock。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, nh=1, ec=128, gc=512, scale=False):
    """Initializes MaxSigmoidAttnBlock with specified arguments."""
    super().__init__()
    self.nh = nh
    self.hc = c2 // nh
    self.ec = Conv(c1, ec, k=1, act=False) if c1 != ec else None
    self.gl = nn.Linear(gc, ec)
    self.bias = nn.Parameter(torch.zeros(nh))
    self.proj_conv = Conv(c1, c2, k=3, s=1, act=False)
    self.scale = nn.Parameter(torch.ones(1, nh, 1, 1)) if scale else 1.0

forward(x, guide)

前进过程。

源代码 ultralytics/nn/modules/block.py
def forward(self, x, guide):
    """Forward process."""
    bs, _, h, w = x.shape

    guide = self.gl(guide)
    guide = guide.view(bs, -1, self.nh, self.hc)
    embed = self.ec(x) if self.ec is not None else x
    embed = embed.view(bs, self.nh, self.hc, h, w)

    aw = torch.einsum("bmchw,bnmc->bmhwn", embed, guide)
    aw = aw.max(dim=-1)[0]
    aw = aw / (self.hc**0.5)
    aw = aw + self.bias[None, :, None, None]
    aw = aw.sigmoid() * self.scale

    x = self.proj_conv(x)
    x = x.view(bs, self.nh, -1, h, w)
    x = x * aw.unsqueeze(2)
    return x.view(bs, -1, h, w)



ultralytics.nn.modules.block.C2fAttn

垒球 Module

带有附加 attn 模块的 C2f 模块。

源代码 ultralytics/nn/modules/block.py
class C2fAttn(nn.Module):
    """C2f module with an additional attn module."""

    def __init__(self, c1, c2, n=1, ec=128, nh=1, gc=512, shortcut=False, g=1, e=0.5):
        """Initialize CSP bottleneck layer with two convolutions with arguments ch_in, ch_out, number, shortcut, groups,
        expansion.
        """
        super().__init__()
        self.c = int(c2 * e)  # hidden channels
        self.cv1 = Conv(c1, 2 * self.c, 1, 1)
        self.cv2 = Conv((3 + n) * self.c, c2, 1)  # optional act=FReLU(c2)
        self.m = nn.ModuleList(Bottleneck(self.c, self.c, shortcut, g, k=((3, 3), (3, 3)), e=1.0) for _ in range(n))
        self.attn = MaxSigmoidAttnBlock(self.c, self.c, gc=gc, ec=ec, nh=nh)

    def forward(self, x, guide):
        """Forward pass through C2f layer."""
        y = list(self.cv1(x).chunk(2, 1))
        y.extend(m(y[-1]) for m in self.m)
        y.append(self.attn(y[-1], guide))
        return self.cv2(torch.cat(y, 1))

    def forward_split(self, x, guide):
        """Forward pass using split() instead of chunk()."""
        y = list(self.cv1(x).split((self.c, self.c), 1))
        y.extend(m(y[-1]) for m in self.m)
        y.append(self.attn(y[-1], guide))
        return self.cv2(torch.cat(y, 1))

__init__(c1, c2, n=1, ec=128, nh=1, gc=512, shortcut=False, g=1, e=0.5)

用两个卷积初始化 CSP 瓶颈层,参数为 ch_in、ch_out、number、shortcut、group、 扩展。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, ec=128, nh=1, gc=512, shortcut=False, g=1, e=0.5):
    """Initialize CSP bottleneck layer with two convolutions with arguments ch_in, ch_out, number, shortcut, groups,
    expansion.
    """
    super().__init__()
    self.c = int(c2 * e)  # hidden channels
    self.cv1 = Conv(c1, 2 * self.c, 1, 1)
    self.cv2 = Conv((3 + n) * self.c, c2, 1)  # optional act=FReLU(c2)
    self.m = nn.ModuleList(Bottleneck(self.c, self.c, shortcut, g, k=((3, 3), (3, 3)), e=1.0) for _ in range(n))
    self.attn = MaxSigmoidAttnBlock(self.c, self.c, gc=gc, ec=ec, nh=nh)

forward(x, guide)

向前穿过 C2f 层。

源代码 ultralytics/nn/modules/block.py
def forward(self, x, guide):
    """Forward pass through C2f layer."""
    y = list(self.cv1(x).chunk(2, 1))
    y.extend(m(y[-1]) for m in self.m)
    y.append(self.attn(y[-1], guide))
    return self.cv2(torch.cat(y, 1))

forward_split(x, guide)

使用 split() 代替 chunk() 向前传递。

源代码 ultralytics/nn/modules/block.py
def forward_split(self, x, guide):
    """Forward pass using split() instead of chunk()."""
    y = list(self.cv1(x).split((self.c, self.c), 1))
    y.extend(m(y[-1]) for m in self.m)
    y.append(self.attn(y[-1], guide))
    return self.cv2(torch.cat(y, 1))



ultralytics.nn.modules.block.ImagePoolingAttn

垒球 Module

ImagePoolingAttn:利用图像识别信息增强文本嵌入。

源代码 ultralytics/nn/modules/block.py
class ImagePoolingAttn(nn.Module):
    """ImagePoolingAttn: Enhance the text embeddings with image-aware information."""

    def __init__(self, ec=256, ch=(), ct=512, nh=8, k=3, scale=False):
        """Initializes ImagePoolingAttn with specified arguments."""
        super().__init__()

        nf = len(ch)
        self.query = nn.Sequential(nn.LayerNorm(ct), nn.Linear(ct, ec))
        self.key = nn.Sequential(nn.LayerNorm(ec), nn.Linear(ec, ec))
        self.value = nn.Sequential(nn.LayerNorm(ec), nn.Linear(ec, ec))
        self.proj = nn.Linear(ec, ct)
        self.scale = nn.Parameter(torch.tensor([0.0]), requires_grad=True) if scale else 1.0
        self.projections = nn.ModuleList([nn.Conv2d(in_channels, ec, kernel_size=1) for in_channels in ch])
        self.im_pools = nn.ModuleList([nn.AdaptiveMaxPool2d((k, k)) for _ in range(nf)])
        self.ec = ec
        self.nh = nh
        self.nf = nf
        self.hc = ec // nh
        self.k = k

    def forward(self, x, text):
        """Executes attention mechanism on input tensor x and guide tensor."""
        bs = x[0].shape[0]
        assert len(x) == self.nf
        num_patches = self.k**2
        x = [pool(proj(x)).view(bs, -1, num_patches) for (x, proj, pool) in zip(x, self.projections, self.im_pools)]
        x = torch.cat(x, dim=-1).transpose(1, 2)
        q = self.query(text)
        k = self.key(x)
        v = self.value(x)

        # q = q.reshape(1, text.shape[1], self.nh, self.hc).repeat(bs, 1, 1, 1)
        q = q.reshape(bs, -1, self.nh, self.hc)
        k = k.reshape(bs, -1, self.nh, self.hc)
        v = v.reshape(bs, -1, self.nh, self.hc)

        aw = torch.einsum("bnmc,bkmc->bmnk", q, k)
        aw = aw / (self.hc**0.5)
        aw = F.softmax(aw, dim=-1)

        x = torch.einsum("bmnk,bkmc->bnmc", aw, v)
        x = self.proj(x.reshape(bs, -1, self.ec))
        return x * self.scale + text

__init__(ec=256, ch=(), ct=512, nh=8, k=3, scale=False)

使用指定参数初始化 ImagePoolingAttn。

源代码 ultralytics/nn/modules/block.py
def __init__(self, ec=256, ch=(), ct=512, nh=8, k=3, scale=False):
    """Initializes ImagePoolingAttn with specified arguments."""
    super().__init__()

    nf = len(ch)
    self.query = nn.Sequential(nn.LayerNorm(ct), nn.Linear(ct, ec))
    self.key = nn.Sequential(nn.LayerNorm(ec), nn.Linear(ec, ec))
    self.value = nn.Sequential(nn.LayerNorm(ec), nn.Linear(ec, ec))
    self.proj = nn.Linear(ec, ct)
    self.scale = nn.Parameter(torch.tensor([0.0]), requires_grad=True) if scale else 1.0
    self.projections = nn.ModuleList([nn.Conv2d(in_channels, ec, kernel_size=1) for in_channels in ch])
    self.im_pools = nn.ModuleList([nn.AdaptiveMaxPool2d((k, k)) for _ in range(nf)])
    self.ec = ec
    self.nh = nh
    self.nf = nf
    self.hc = ec // nh
    self.k = k

forward(x, text)

对输入tensor x 和引导tensor 执行关注机制。

源代码 ultralytics/nn/modules/block.py
def forward(self, x, text):
    """Executes attention mechanism on input tensor x and guide tensor."""
    bs = x[0].shape[0]
    assert len(x) == self.nf
    num_patches = self.k**2
    x = [pool(proj(x)).view(bs, -1, num_patches) for (x, proj, pool) in zip(x, self.projections, self.im_pools)]
    x = torch.cat(x, dim=-1).transpose(1, 2)
    q = self.query(text)
    k = self.key(x)
    v = self.value(x)

    # q = q.reshape(1, text.shape[1], self.nh, self.hc).repeat(bs, 1, 1, 1)
    q = q.reshape(bs, -1, self.nh, self.hc)
    k = k.reshape(bs, -1, self.nh, self.hc)
    v = v.reshape(bs, -1, self.nh, self.hc)

    aw = torch.einsum("bnmc,bkmc->bmnk", q, k)
    aw = aw / (self.hc**0.5)
    aw = F.softmax(aw, dim=-1)

    x = torch.einsum("bmnk,bkmc->bnmc", aw, v)
    x = self.proj(x.reshape(bs, -1, self.ec))
    return x * self.scale + text



ultralytics.nn.modules.block.ContrastiveHead

垒球 Module

Contrastive Head forYOLO-World 根据图像和文本特征之间的相似性计算区域-文本分数。 特征。

源代码 ultralytics/nn/modules/block.py
class ContrastiveHead(nn.Module):
    """Contrastive Head for YOLO-World compute the region-text scores according to the similarity between image and text
    features.
    """

    def __init__(self):
        """Initializes ContrastiveHead with specified region-text similarity parameters."""
        super().__init__()
        # NOTE: use -10.0 to keep the init cls loss consistency with other losses
        self.bias = nn.Parameter(torch.tensor([-10.0]))
        self.logit_scale = nn.Parameter(torch.ones([]) * torch.tensor(1 / 0.07).log())

    def forward(self, x, w):
        """Forward function of contrastive learning."""
        x = F.normalize(x, dim=1, p=2)
        w = F.normalize(w, dim=-1, p=2)
        x = torch.einsum("bchw,bkc->bkhw", x, w)
        return x * self.logit_scale.exp() + self.bias

__init__()

使用指定的区域-文本相似性参数初始化 ContrastiveHead。

源代码 ultralytics/nn/modules/block.py
def __init__(self):
    """Initializes ContrastiveHead with specified region-text similarity parameters."""
    super().__init__()
    # NOTE: use -10.0 to keep the init cls loss consistency with other losses
    self.bias = nn.Parameter(torch.tensor([-10.0]))
    self.logit_scale = nn.Parameter(torch.ones([]) * torch.tensor(1 / 0.07).log())

forward(x, w)

对比学习的前向功能

源代码 ultralytics/nn/modules/block.py
def forward(self, x, w):
    """Forward function of contrastive learning."""
    x = F.normalize(x, dim=1, p=2)
    w = F.normalize(w, dim=-1, p=2)
    x = torch.einsum("bchw,bkc->bkhw", x, w)
    return x * self.logit_scale.exp() + self.bias



ultralytics.nn.modules.block.BNContrastiveHead

垒球 Module

YOLO-World 的批量规范对比头,使用批量规范而不是 l2 规范化。

参数

名称 类型 说明 默认值
embed_dims int

嵌入文本和图像特征的尺寸。

所需
源代码 ultralytics/nn/modules/block.py
class BNContrastiveHead(nn.Module):
    """
    Batch Norm Contrastive Head for YOLO-World using batch norm instead of l2-normalization.

    Args:
        embed_dims (int): Embed dimensions of text and image features.
    """

    def __init__(self, embed_dims: int):
        """Initialize ContrastiveHead with region-text similarity parameters."""
        super().__init__()
        self.norm = nn.BatchNorm2d(embed_dims)
        # NOTE: use -10.0 to keep the init cls loss consistency with other losses
        self.bias = nn.Parameter(torch.tensor([-10.0]))
        # use -1.0 is more stable
        self.logit_scale = nn.Parameter(-1.0 * torch.ones([]))

    def forward(self, x, w):
        """Forward function of contrastive learning."""
        x = self.norm(x)
        w = F.normalize(w, dim=-1, p=2)
        x = torch.einsum("bchw,bkc->bkhw", x, w)
        return x * self.logit_scale.exp() + self.bias

__init__(embed_dims)

使用区域-文本相似性参数初始化 ContrastiveHead。

源代码 ultralytics/nn/modules/block.py
def __init__(self, embed_dims: int):
    """Initialize ContrastiveHead with region-text similarity parameters."""
    super().__init__()
    self.norm = nn.BatchNorm2d(embed_dims)
    # NOTE: use -10.0 to keep the init cls loss consistency with other losses
    self.bias = nn.Parameter(torch.tensor([-10.0]))
    # use -1.0 is more stable
    self.logit_scale = nn.Parameter(-1.0 * torch.ones([]))

forward(x, w)

对比学习的前向功能

源代码 ultralytics/nn/modules/block.py
def forward(self, x, w):
    """Forward function of contrastive learning."""
    x = self.norm(x)
    w = F.normalize(w, dim=-1, p=2)
    x = torch.einsum("bchw,bkc->bkhw", x, w)
    return x * self.logit_scale.exp() + self.bias



ultralytics.nn.modules.block.RepBottleneck

垒球 Bottleneck

代表瓶颈。

源代码 ultralytics/nn/modules/block.py
class RepBottleneck(Bottleneck):
    """Rep bottleneck."""

    def __init__(self, c1, c2, shortcut=True, g=1, k=(3, 3), e=0.5):
        """Initializes a RepBottleneck module with customizable in/out channels, shortcut option, groups and expansion
        ratio.
        """
        super().__init__(c1, c2, shortcut, g, k, e)
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = RepConv(c1, c_, k[0], 1)

__init__(c1, c2, shortcut=True, g=1, k=(3, 3), e=0.5)

初始化 RepBottleneck 模块,该模块具有可定制的输入/输出通道、快捷选项、组和扩展比例。 比率。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, shortcut=True, g=1, k=(3, 3), e=0.5):
    """Initializes a RepBottleneck module with customizable in/out channels, shortcut option, groups and expansion
    ratio.
    """
    super().__init__(c1, c2, shortcut, g, k, e)
    c_ = int(c2 * e)  # hidden channels
    self.cv1 = RepConv(c1, c_, k[0], 1)



ultralytics.nn.modules.block.RepCSP

垒球 C3

Rep CSP Bottleneck with 3 convolutions.

源代码 ultralytics/nn/modules/block.py
class RepCSP(C3):
    """Rep CSP Bottleneck with 3 convolutions."""

    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
        """Initializes RepCSP layer with given channels, repetitions, shortcut, groups and expansion ratio."""
        super().__init__(c1, c2, n, shortcut, g, e)
        c_ = int(c2 * e)  # hidden channels
        self.m = nn.Sequential(*(RepBottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))

__init__(c1, c2, n=1, shortcut=True, g=1, e=0.5)

用给定的通道、重复次数、快捷方式、组和扩展率初始化 RepCSP 层。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):
    """Initializes RepCSP layer with given channels, repetitions, shortcut, groups and expansion ratio."""
    super().__init__(c1, c2, n, shortcut, g, e)
    c_ = int(c2 * e)  # hidden channels
    self.m = nn.Sequential(*(RepBottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))



ultralytics.nn.modules.block.RepNCSPELAN4

垒球 Module

CSP-ELAN.

源代码 ultralytics/nn/modules/block.py
class RepNCSPELAN4(nn.Module):
    """CSP-ELAN."""

    def __init__(self, c1, c2, c3, c4, n=1):
        """Initializes CSP-ELAN layer with specified channel sizes, repetitions, and convolutions."""
        super().__init__()
        self.c = c3 // 2
        self.cv1 = Conv(c1, c3, 1, 1)
        self.cv2 = nn.Sequential(RepCSP(c3 // 2, c4, n), Conv(c4, c4, 3, 1))
        self.cv3 = nn.Sequential(RepCSP(c4, c4, n), Conv(c4, c4, 3, 1))
        self.cv4 = Conv(c3 + (2 * c4), c2, 1, 1)

    def forward(self, x):
        """Forward pass through RepNCSPELAN4 layer."""
        y = list(self.cv1(x).chunk(2, 1))
        y.extend((m(y[-1])) for m in [self.cv2, self.cv3])
        return self.cv4(torch.cat(y, 1))

    def forward_split(self, x):
        """Forward pass using split() instead of chunk()."""
        y = list(self.cv1(x).split((self.c, self.c), 1))
        y.extend(m(y[-1]) for m in [self.cv2, self.cv3])
        return self.cv4(torch.cat(y, 1))

__init__(c1, c2, c3, c4, n=1)

用指定的通道大小、重复次数和卷积来初始化 CSP-ELAN 层。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, c3, c4, n=1):
    """Initializes CSP-ELAN layer with specified channel sizes, repetitions, and convolutions."""
    super().__init__()
    self.c = c3 // 2
    self.cv1 = Conv(c1, c3, 1, 1)
    self.cv2 = nn.Sequential(RepCSP(c3 // 2, c4, n), Conv(c4, c4, 3, 1))
    self.cv3 = nn.Sequential(RepCSP(c4, c4, n), Conv(c4, c4, 3, 1))
    self.cv4 = Conv(c3 + (2 * c4), c2, 1, 1)

forward(x)

前向通过 RepNCSPELAN4 层。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through RepNCSPELAN4 layer."""
    y = list(self.cv1(x).chunk(2, 1))
    y.extend((m(y[-1])) for m in [self.cv2, self.cv3])
    return self.cv4(torch.cat(y, 1))

forward_split(x)

使用 split() 代替 chunk() 向前传递。

源代码 ultralytics/nn/modules/block.py
def forward_split(self, x):
    """Forward pass using split() instead of chunk()."""
    y = list(self.cv1(x).split((self.c, self.c), 1))
    y.extend(m(y[-1]) for m in [self.cv2, self.cv3])
    return self.cv4(torch.cat(y, 1))



ultralytics.nn.modules.block.ELAN1

垒球 RepNCSPELAN4

ELAN1 module with 4 convolutions.

源代码 ultralytics/nn/modules/block.py
class ELAN1(RepNCSPELAN4):
    """ELAN1 module with 4 convolutions."""

    def __init__(self, c1, c2, c3, c4):
        """Initializes ELAN1 layer with specified channel sizes."""
        super().__init__(c1, c2, c3, c4)
        self.c = c3 // 2
        self.cv1 = Conv(c1, c3, 1, 1)
        self.cv2 = Conv(c3 // 2, c4, 3, 1)
        self.cv3 = Conv(c4, c4, 3, 1)
        self.cv4 = Conv(c3 + (2 * c4), c2, 1, 1)

__init__(c1, c2, c3, c4)

Initializes ELAN1 layer with specified channel sizes.

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, c3, c4):
    """Initializes ELAN1 layer with specified channel sizes."""
    super().__init__(c1, c2, c3, c4)
    self.c = c3 // 2
    self.cv1 = Conv(c1, c3, 1, 1)
    self.cv2 = Conv(c3 // 2, c4, 3, 1)
    self.cv3 = Conv(c4, c4, 3, 1)
    self.cv4 = Conv(c3 + (2 * c4), c2, 1, 1)



ultralytics.nn.modules.block.AConv

垒球 Module

AConv.

源代码 ultralytics/nn/modules/block.py
class AConv(nn.Module):
    """AConv."""

    def __init__(self, c1, c2):
        """Initializes AConv module with convolution layers."""
        super().__init__()
        self.cv1 = Conv(c1, c2, 3, 2, 1)

    def forward(self, x):
        """Forward pass through AConv layer."""
        x = torch.nn.functional.avg_pool2d(x, 2, 1, 0, False, True)
        return self.cv1(x)

__init__(c1, c2)

Initializes AConv module with convolution layers.

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2):
    """Initializes AConv module with convolution layers."""
    super().__init__()
    self.cv1 = Conv(c1, c2, 3, 2, 1)

forward(x)

Forward pass through AConv layer.

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through AConv layer."""
    x = torch.nn.functional.avg_pool2d(x, 2, 1, 0, False, True)
    return self.cv1(x)



ultralytics.nn.modules.block.ADown

垒球 Module

ADown.

源代码 ultralytics/nn/modules/block.py
class ADown(nn.Module):
    """ADown."""

    def __init__(self, c1, c2):
        """Initializes ADown module with convolution layers to downsample input from channels c1 to c2."""
        super().__init__()
        self.c = c2 // 2
        self.cv1 = Conv(c1 // 2, self.c, 3, 2, 1)
        self.cv2 = Conv(c1 // 2, self.c, 1, 1, 0)

    def forward(self, x):
        """Forward pass through ADown layer."""
        x = torch.nn.functional.avg_pool2d(x, 2, 1, 0, False, True)
        x1, x2 = x.chunk(2, 1)
        x1 = self.cv1(x1)
        x2 = torch.nn.functional.max_pool2d(x2, 3, 2, 1)
        x2 = self.cv2(x2)
        return torch.cat((x1, x2), 1)

__init__(c1, c2)

初始化带有卷积层的 ADown 模块,对来自通道 c1 至 c2 的输入进行降采样。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2):
    """Initializes ADown module with convolution layers to downsample input from channels c1 to c2."""
    super().__init__()
    self.c = c2 // 2
    self.cv1 = Conv(c1 // 2, self.c, 3, 2, 1)
    self.cv2 = Conv(c1 // 2, self.c, 1, 1, 0)

forward(x)

前向通过 ADown 层。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through ADown layer."""
    x = torch.nn.functional.avg_pool2d(x, 2, 1, 0, False, True)
    x1, x2 = x.chunk(2, 1)
    x1 = self.cv1(x1)
    x2 = torch.nn.functional.max_pool2d(x2, 3, 2, 1)
    x2 = self.cv2(x2)
    return torch.cat((x1, x2), 1)



ultralytics.nn.modules.block.SPPELAN

垒球 Module

SPP-ELAN.

源代码 ultralytics/nn/modules/block.py
class SPPELAN(nn.Module):
    """SPP-ELAN."""

    def __init__(self, c1, c2, c3, k=5):
        """Initializes SPP-ELAN block with convolution and max pooling layers for spatial pyramid pooling."""
        super().__init__()
        self.c = c3
        self.cv1 = Conv(c1, c3, 1, 1)
        self.cv2 = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)
        self.cv3 = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)
        self.cv4 = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)
        self.cv5 = Conv(4 * c3, c2, 1, 1)

    def forward(self, x):
        """Forward pass through SPPELAN layer."""
        y = [self.cv1(x)]
        y.extend(m(y[-1]) for m in [self.cv2, self.cv3, self.cv4])
        return self.cv5(torch.cat(y, 1))

__init__(c1, c2, c3, k=5)

初始化 SPP-ELAN 块的卷积层和最大池化层,用于空间金字塔池化。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, c3, k=5):
    """Initializes SPP-ELAN block with convolution and max pooling layers for spatial pyramid pooling."""
    super().__init__()
    self.c = c3
    self.cv1 = Conv(c1, c3, 1, 1)
    self.cv2 = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)
    self.cv3 = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)
    self.cv4 = nn.MaxPool2d(kernel_size=k, stride=1, padding=k // 2)
    self.cv5 = Conv(4 * c3, c2, 1, 1)

forward(x)

前向通过 SPPELAN 层。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through SPPELAN layer."""
    y = [self.cv1(x)]
    y.extend(m(y[-1]) for m in [self.cv2, self.cv3, self.cv4])
    return self.cv5(torch.cat(y, 1))



ultralytics.nn.modules.block.CBLinear

垒球 Module

CBLinear.

源代码 ultralytics/nn/modules/block.py
class CBLinear(nn.Module):
    """CBLinear."""

    def __init__(self, c1, c2s, k=1, s=1, p=None, g=1):
        """Initializes the CBLinear module, passing inputs unchanged."""
        super(CBLinear, self).__init__()
        self.c2s = c2s
        self.conv = nn.Conv2d(c1, sum(c2s), k, s, autopad(k, p), groups=g, bias=True)

    def forward(self, x):
        """Forward pass through CBLinear layer."""
        return self.conv(x).split(self.c2s, dim=1)

__init__(c1, c2s, k=1, s=1, p=None, g=1)

初始化 CBLinear 模块,输入不变。

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2s, k=1, s=1, p=None, g=1):
    """Initializes the CBLinear module, passing inputs unchanged."""
    super(CBLinear, self).__init__()
    self.c2s = c2s
    self.conv = nn.Conv2d(c1, sum(c2s), k, s, autopad(k, p), groups=g, bias=True)

forward(x)

向前通过 CBLinear 层。

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """Forward pass through CBLinear layer."""
    return self.conv(x).split(self.c2s, dim=1)



ultralytics.nn.modules.block.CBFuse

垒球 Module

CBFuse.

源代码 ultralytics/nn/modules/block.py
class CBFuse(nn.Module):
    """CBFuse."""

    def __init__(self, idx):
        """Initializes CBFuse module with layer index for selective feature fusion."""
        super(CBFuse, self).__init__()
        self.idx = idx

    def forward(self, xs):
        """Forward pass through CBFuse layer."""
        target_size = xs[-1].shape[2:]
        res = [F.interpolate(x[self.idx[i]], size=target_size, mode="nearest") for i, x in enumerate(xs[:-1])]
        return torch.sum(torch.stack(res + xs[-1:]), dim=0)

__init__(idx)

使用层索引初始化 CBFuse 模块,以进行选择性特征融合。

源代码 ultralytics/nn/modules/block.py
def __init__(self, idx):
    """Initializes CBFuse module with layer index for selective feature fusion."""
    super(CBFuse, self).__init__()
    self.idx = idx

forward(xs)

正向通过 CBFuse 层。

源代码 ultralytics/nn/modules/block.py
def forward(self, xs):
    """Forward pass through CBFuse layer."""
    target_size = xs[-1].shape[2:]
    res = [F.interpolate(x[self.idx[i]], size=target_size, mode="nearest") for i, x in enumerate(xs[:-1])]
    return torch.sum(torch.stack(res + xs[-1:]), dim=0)



ultralytics.nn.modules.block.RepVGGDW

垒球 Module

RepVGGDW is a class that represents a depth wise separable convolutional block in RepVGG architecture.

源代码 ultralytics/nn/modules/block.py
class RepVGGDW(torch.nn.Module):
    """RepVGGDW is a class that represents a depth wise separable convolutional block in RepVGG architecture."""

    def __init__(self, ed) -> None:
        super().__init__()
        self.conv = Conv(ed, ed, 7, 1, 3, g=ed, act=False)
        self.conv1 = Conv(ed, ed, 3, 1, 1, g=ed, act=False)
        self.dim = ed
        self.act = nn.SiLU()

    def forward(self, x):
        """
        Performs a forward pass of the RepVGGDW block.

        Args:
            x (torch.Tensor): Input tensor.

        Returns:
            (torch.Tensor): Output tensor after applying the depth wise separable convolution.
        """
        return self.act(self.conv(x) + self.conv1(x))

    def forward_fuse(self, x):
        """
        Performs a forward pass of the RepVGGDW block without fusing the convolutions.

        Args:
            x (torch.Tensor): Input tensor.

        Returns:
            (torch.Tensor): Output tensor after applying the depth wise separable convolution.
        """
        return self.act(self.conv(x))

    @torch.no_grad()
    def fuse(self):
        """
        Fuses the convolutional layers in the RepVGGDW block.

        This method fuses the convolutional layers and updates the weights and biases accordingly.
        """
        conv = fuse_conv_and_bn(self.conv.conv, self.conv.bn)
        conv1 = fuse_conv_and_bn(self.conv1.conv, self.conv1.bn)

        conv_w = conv.weight
        conv_b = conv.bias
        conv1_w = conv1.weight
        conv1_b = conv1.bias

        conv1_w = torch.nn.functional.pad(conv1_w, [2, 2, 2, 2])

        final_conv_w = conv_w + conv1_w
        final_conv_b = conv_b + conv1_b

        conv.weight.data.copy_(final_conv_w)
        conv.bias.data.copy_(final_conv_b)

        self.conv = conv
        del self.conv1

forward(x)

Performs a forward pass of the RepVGGDW block.

参数

名称 类型 说明 默认值
x Tensor

Input tensor.

所需

返回:

类型 说明
Tensor

Output tensor after applying the depth wise separable convolution.

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """
    Performs a forward pass of the RepVGGDW block.

    Args:
        x (torch.Tensor): Input tensor.

    Returns:
        (torch.Tensor): Output tensor after applying the depth wise separable convolution.
    """
    return self.act(self.conv(x) + self.conv1(x))

forward_fuse(x)

Performs a forward pass of the RepVGGDW block without fusing the convolutions.

参数

名称 类型 说明 默认值
x Tensor

Input tensor.

所需

返回:

类型 说明
Tensor

Output tensor after applying the depth wise separable convolution.

源代码 ultralytics/nn/modules/block.py
def forward_fuse(self, x):
    """
    Performs a forward pass of the RepVGGDW block without fusing the convolutions.

    Args:
        x (torch.Tensor): Input tensor.

    Returns:
        (torch.Tensor): Output tensor after applying the depth wise separable convolution.
    """
    return self.act(self.conv(x))

fuse()

Fuses the convolutional layers in the RepVGGDW block.

This method fuses the convolutional layers and updates the weights and biases accordingly.

源代码 ultralytics/nn/modules/block.py
@torch.no_grad()
def fuse(self):
    """
    Fuses the convolutional layers in the RepVGGDW block.

    This method fuses the convolutional layers and updates the weights and biases accordingly.
    """
    conv = fuse_conv_and_bn(self.conv.conv, self.conv.bn)
    conv1 = fuse_conv_and_bn(self.conv1.conv, self.conv1.bn)

    conv_w = conv.weight
    conv_b = conv.bias
    conv1_w = conv1.weight
    conv1_b = conv1.bias

    conv1_w = torch.nn.functional.pad(conv1_w, [2, 2, 2, 2])

    final_conv_w = conv_w + conv1_w
    final_conv_b = conv_b + conv1_b

    conv.weight.data.copy_(final_conv_w)
    conv.bias.data.copy_(final_conv_b)

    self.conv = conv
    del self.conv1



ultralytics.nn.modules.block.CIB

垒球 Module

Conditional Identity Block (CIB) module.

参数

名称 类型 说明 默认值
c1 int

输入通道数

所需
c2 int

Number of output channels.

所需
shortcut bool

Whether to add a shortcut connection. Defaults to True.

True
e float

Scaling factor for the hidden channels. Defaults to 0.5.

0.5
lk bool

Whether to use RepVGGDW for the third convolutional layer. Defaults to False.

False
源代码 ultralytics/nn/modules/block.py
class CIB(nn.Module):
    """
    Conditional Identity Block (CIB) module.

    Args:
        c1 (int): Number of input channels.
        c2 (int): Number of output channels.
        shortcut (bool, optional): Whether to add a shortcut connection. Defaults to True.
        e (float, optional): Scaling factor for the hidden channels. Defaults to 0.5.
        lk (bool, optional): Whether to use RepVGGDW for the third convolutional layer. Defaults to False.
    """

    def __init__(self, c1, c2, shortcut=True, e=0.5, lk=False):
        """Initializes the custom model with optional shortcut, scaling factor, and RepVGGDW layer."""
        super().__init__()
        c_ = int(c2 * e)  # hidden channels
        self.cv1 = nn.Sequential(
            Conv(c1, c1, 3, g=c1),
            Conv(c1, 2 * c_, 1),
            RepVGGDW(2 * c_) if lk else Conv(2 * c_, 2 * c_, 3, g=2 * c_),
            Conv(2 * c_, c2, 1),
            Conv(c2, c2, 3, g=c2),
        )

        self.add = shortcut and c1 == c2

    def forward(self, x):
        """
        Forward pass of the CIB module.

        Args:
            x (torch.Tensor): Input tensor.

        Returns:
            (torch.Tensor): Output tensor.
        """
        return x + self.cv1(x) if self.add else self.cv1(x)

__init__(c1, c2, shortcut=True, e=0.5, lk=False)

Initializes the custom model with optional shortcut, scaling factor, and RepVGGDW layer.

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, shortcut=True, e=0.5, lk=False):
    """Initializes the custom model with optional shortcut, scaling factor, and RepVGGDW layer."""
    super().__init__()
    c_ = int(c2 * e)  # hidden channels
    self.cv1 = nn.Sequential(
        Conv(c1, c1, 3, g=c1),
        Conv(c1, 2 * c_, 1),
        RepVGGDW(2 * c_) if lk else Conv(2 * c_, 2 * c_, 3, g=2 * c_),
        Conv(2 * c_, c2, 1),
        Conv(c2, c2, 3, g=c2),
    )

    self.add = shortcut and c1 == c2

forward(x)

Forward pass of the CIB module.

参数

名称 类型 说明 默认值
x Tensor

Input tensor.

所需

返回:

类型 说明
Tensor

Output tensor.

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """
    Forward pass of the CIB module.

    Args:
        x (torch.Tensor): Input tensor.

    Returns:
        (torch.Tensor): Output tensor.
    """
    return x + self.cv1(x) if self.add else self.cv1(x)



ultralytics.nn.modules.block.C2fCIB

垒球 C2f

C2fCIB class represents a convolutional block with C2f and CIB modules.

参数

名称 类型 说明 默认值
c1 int

输入通道数

所需
c2 int

Number of output channels.

所需
n int

Number of CIB modules to stack. Defaults to 1.

1
shortcut bool

Whether to use shortcut connection. Defaults to False.

False
lk bool

Whether to use local key connection. Defaults to False.

False
g int

Number of groups for grouped convolution. Defaults to 1.

1
e float

Expansion ratio for CIB modules. Defaults to 0.5.

0.5
源代码 ultralytics/nn/modules/block.py
class C2fCIB(C2f):
    """
    C2fCIB class represents a convolutional block with C2f and CIB modules.

    Args:
        c1 (int): Number of input channels.
        c2 (int): Number of output channels.
        n (int, optional): Number of CIB modules to stack. Defaults to 1.
        shortcut (bool, optional): Whether to use shortcut connection. Defaults to False.
        lk (bool, optional): Whether to use local key connection. Defaults to False.
        g (int, optional): Number of groups for grouped convolution. Defaults to 1.
        e (float, optional): Expansion ratio for CIB modules. Defaults to 0.5.
    """

    def __init__(self, c1, c2, n=1, shortcut=False, lk=False, g=1, e=0.5):
        """Initializes the module with specified parameters for channel, shortcut, local key, groups, and expansion."""
        super().__init__(c1, c2, n, shortcut, g, e)
        self.m = nn.ModuleList(CIB(self.c, self.c, shortcut, e=1.0, lk=lk) for _ in range(n))

__init__(c1, c2, n=1, shortcut=False, lk=False, g=1, e=0.5)

Initializes the module with specified parameters for channel, shortcut, local key, groups, and expansion.

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, n=1, shortcut=False, lk=False, g=1, e=0.5):
    """Initializes the module with specified parameters for channel, shortcut, local key, groups, and expansion."""
    super().__init__(c1, c2, n, shortcut, g, e)
    self.m = nn.ModuleList(CIB(self.c, self.c, shortcut, e=1.0, lk=lk) for _ in range(n))



ultralytics.nn.modules.block.Attention

垒球 Module

Attention module that performs self-attention on the input tensor.

参数

名称 类型 说明 默认值
dim int

The input tensor dimension.

所需
num_heads int

注意头的数量。

8
attn_ratio float

The ratio of the attention key dimension to the head dimension.

0.5

属性

名称 类型 说明
num_heads int

注意头的数量。

head_dim int

The dimension of each attention head.

key_dim int

The dimension of the attention key.

scale float

The scaling factor for the attention scores.

qkv Conv

Convolutional layer for computing the query, key, and value.

proj Conv

Convolutional layer for projecting the attended values.

pe Conv

Convolutional layer for positional encoding.

源代码 ultralytics/nn/modules/block.py
class Attention(nn.Module):
    """
    Attention module that performs self-attention on the input tensor.

    Args:
        dim (int): The input tensor dimension.
        num_heads (int): The number of attention heads.
        attn_ratio (float): The ratio of the attention key dimension to the head dimension.

    Attributes:
        num_heads (int): The number of attention heads.
        head_dim (int): The dimension of each attention head.
        key_dim (int): The dimension of the attention key.
        scale (float): The scaling factor for the attention scores.
        qkv (Conv): Convolutional layer for computing the query, key, and value.
        proj (Conv): Convolutional layer for projecting the attended values.
        pe (Conv): Convolutional layer for positional encoding.
    """

    def __init__(self, dim, num_heads=8, attn_ratio=0.5):
        """Initializes multi-head attention module with query, key, and value convolutions and positional encoding."""
        super().__init__()
        self.num_heads = num_heads
        self.head_dim = dim // num_heads
        self.key_dim = int(self.head_dim * attn_ratio)
        self.scale = self.key_dim**-0.5
        nh_kd = nh_kd = self.key_dim * num_heads
        h = dim + nh_kd * 2
        self.qkv = Conv(dim, h, 1, act=False)
        self.proj = Conv(dim, dim, 1, act=False)
        self.pe = Conv(dim, dim, 3, 1, g=dim, act=False)

    def forward(self, x):
        """
        Forward pass of the Attention module.

        Args:
            x (torch.Tensor): The input tensor.

        Returns:
            (torch.Tensor): The output tensor after self-attention.
        """
        B, C, H, W = x.shape
        N = H * W
        qkv = self.qkv(x)
        q, k, v = qkv.view(B, self.num_heads, self.key_dim * 2 + self.head_dim, N).split(
            [self.key_dim, self.key_dim, self.head_dim], dim=2
        )

        attn = (q.transpose(-2, -1) @ k) * self.scale
        attn = attn.softmax(dim=-1)
        x = (v @ attn.transpose(-2, -1)).view(B, C, H, W) + self.pe(v.reshape(B, C, H, W))
        x = self.proj(x)
        return x

__init__(dim, num_heads=8, attn_ratio=0.5)

Initializes multi-head attention module with query, key, and value convolutions and positional encoding.

源代码 ultralytics/nn/modules/block.py
def __init__(self, dim, num_heads=8, attn_ratio=0.5):
    """Initializes multi-head attention module with query, key, and value convolutions and positional encoding."""
    super().__init__()
    self.num_heads = num_heads
    self.head_dim = dim // num_heads
    self.key_dim = int(self.head_dim * attn_ratio)
    self.scale = self.key_dim**-0.5
    nh_kd = nh_kd = self.key_dim * num_heads
    h = dim + nh_kd * 2
    self.qkv = Conv(dim, h, 1, act=False)
    self.proj = Conv(dim, dim, 1, act=False)
    self.pe = Conv(dim, dim, 3, 1, g=dim, act=False)

forward(x)

Forward pass of the Attention module.

参数

名称 类型 说明 默认值
x Tensor

输入tensor 。

所需

返回:

类型 说明
Tensor

The output tensor after self-attention.

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """
    Forward pass of the Attention module.

    Args:
        x (torch.Tensor): The input tensor.

    Returns:
        (torch.Tensor): The output tensor after self-attention.
    """
    B, C, H, W = x.shape
    N = H * W
    qkv = self.qkv(x)
    q, k, v = qkv.view(B, self.num_heads, self.key_dim * 2 + self.head_dim, N).split(
        [self.key_dim, self.key_dim, self.head_dim], dim=2
    )

    attn = (q.transpose(-2, -1) @ k) * self.scale
    attn = attn.softmax(dim=-1)
    x = (v @ attn.transpose(-2, -1)).view(B, C, H, W) + self.pe(v.reshape(B, C, H, W))
    x = self.proj(x)
    return x



ultralytics.nn.modules.block.PSA

垒球 Module

Position-wise Spatial Attention module.

参数

名称 类型 说明 默认值
c1 int

输入通道数

所需
c2 int

Number of output channels.

所需
e float

Expansion factor for the intermediate channels. Default is 0.5.

0.5

属性

名称 类型 说明
c int

Number of intermediate channels.

cv1 Conv

1x1 convolution layer to reduce the number of input channels to 2*c.

cv2 Conv

1x1 convolution layer to reduce the number of output channels to c.

attn Attention

Attention module for spatial attention.

ffn Sequential

Feed-forward network module.

源代码 ultralytics/nn/modules/block.py
class PSA(nn.Module):
    """
    Position-wise Spatial Attention module.

    Args:
        c1 (int): Number of input channels.
        c2 (int): Number of output channels.
        e (float): Expansion factor for the intermediate channels. Default is 0.5.

    Attributes:
        c (int): Number of intermediate channels.
        cv1 (Conv): 1x1 convolution layer to reduce the number of input channels to 2*c.
        cv2 (Conv): 1x1 convolution layer to reduce the number of output channels to c.
        attn (Attention): Attention module for spatial attention.
        ffn (nn.Sequential): Feed-forward network module.
    """

    def __init__(self, c1, c2, e=0.5):
        """Initializes convolution layers, attention module, and feed-forward network with channel reduction."""
        super().__init__()
        assert c1 == c2
        self.c = int(c1 * e)
        self.cv1 = Conv(c1, 2 * self.c, 1, 1)
        self.cv2 = Conv(2 * self.c, c1, 1)

        self.attn = Attention(self.c, attn_ratio=0.5, num_heads=self.c // 64)
        self.ffn = nn.Sequential(Conv(self.c, self.c * 2, 1), Conv(self.c * 2, self.c, 1, act=False))

    def forward(self, x):
        """
        Forward pass of the PSA module.

        Args:
            x (torch.Tensor): Input tensor.

        Returns:
            (torch.Tensor): Output tensor.
        """
        a, b = self.cv1(x).split((self.c, self.c), dim=1)
        b = b + self.attn(b)
        b = b + self.ffn(b)
        return self.cv2(torch.cat((a, b), 1))

__init__(c1, c2, e=0.5)

Initializes convolution layers, attention module, and feed-forward network with channel reduction.

源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, e=0.5):
    """Initializes convolution layers, attention module, and feed-forward network with channel reduction."""
    super().__init__()
    assert c1 == c2
    self.c = int(c1 * e)
    self.cv1 = Conv(c1, 2 * self.c, 1, 1)
    self.cv2 = Conv(2 * self.c, c1, 1)

    self.attn = Attention(self.c, attn_ratio=0.5, num_heads=self.c // 64)
    self.ffn = nn.Sequential(Conv(self.c, self.c * 2, 1), Conv(self.c * 2, self.c, 1, act=False))

forward(x)

Forward pass of the PSA module.

参数

名称 类型 说明 默认值
x Tensor

Input tensor.

所需

返回:

类型 说明
Tensor

Output tensor.

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """
    Forward pass of the PSA module.

    Args:
        x (torch.Tensor): Input tensor.

    Returns:
        (torch.Tensor): Output tensor.
    """
    a, b = self.cv1(x).split((self.c, self.c), dim=1)
    b = b + self.attn(b)
    b = b + self.ffn(b)
    return self.cv2(torch.cat((a, b), 1))



ultralytics.nn.modules.block.SCDown

垒球 Module

源代码 ultralytics/nn/modules/block.py
class SCDown(nn.Module):
    def __init__(self, c1, c2, k, s):
        """
        Spatial Channel Downsample (SCDown) module.

        Args:
            c1 (int): Number of input channels.
            c2 (int): Number of output channels.
            k (int): Kernel size for the convolutional layer.
            s (int): Stride for the convolutional layer.
        """
        super().__init__()
        self.cv1 = Conv(c1, c2, 1, 1)
        self.cv2 = Conv(c2, c2, k=k, s=s, g=c2, act=False)

    def forward(self, x):
        """
        Forward pass of the SCDown module.

        Args:
            x (torch.Tensor): Input tensor.

        Returns:
            (torch.Tensor): Output tensor after applying the SCDown module.
        """
        return self.cv2(self.cv1(x))

__init__(c1, c2, k, s)

Spatial Channel Downsample (SCDown) module.

参数

名称 类型 说明 默认值
c1 int

输入通道数

所需
c2 int

Number of output channels.

所需
k int

Kernel size for the convolutional layer.

所需
s int

Stride for the convolutional layer.

所需
源代码 ultralytics/nn/modules/block.py
def __init__(self, c1, c2, k, s):
    """
    Spatial Channel Downsample (SCDown) module.

    Args:
        c1 (int): Number of input channels.
        c2 (int): Number of output channels.
        k (int): Kernel size for the convolutional layer.
        s (int): Stride for the convolutional layer.
    """
    super().__init__()
    self.cv1 = Conv(c1, c2, 1, 1)
    self.cv2 = Conv(c2, c2, k=k, s=s, g=c2, act=False)

forward(x)

Forward pass of the SCDown module.

参数

名称 类型 说明 默认值
x Tensor

Input tensor.

所需

返回:

类型 说明
Tensor

Output tensor after applying the SCDown module.

源代码 ultralytics/nn/modules/block.py
def forward(self, x):
    """
    Forward pass of the SCDown module.

    Args:
        x (torch.Tensor): Input tensor.

    Returns:
        (torch.Tensor): Output tensor after applying the SCDown module.
    """
    return self.cv2(self.cv1(x))





Created 2023-11-12, Updated 2024-06-20
Authors: Burhan-Q (2), Laughing-q (3), glenn-jocher (7)