add yolo v10 and modify pipeline

ultralytics/nn/modules/__init__.py

@@ -1,29 +1,147 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
 """
-Ultralytics modules. Visualize with:
+Ultralytics modules.
 
-from ultralytics.nn.modules import *
-import torch
-import os
+Example:
+    Visualize a module with Netron.
+    ```python
+    from ultralytics.nn.modules import *
+    import torch
+    import os
 
-x = torch.ones(1, 128, 40, 40)
-m = Conv(128, 128)
-f = f'{m._get_name()}.onnx'
-torch.onnx.export(m, x, f)
-os.system(f'onnxsim {f} {f} && open {f}')
+    x = torch.ones(1, 128, 40, 40)
+    m = Conv(128, 128)
+    f = f'{m._get_name()}.onnx'
+    torch.onnx.export(m, x, f)
+    os.system(f'onnxslim {f} {f} && open {f}') # pip install onnxslim
+    ```
 """
 
-from .block import (C1, C2, C3, C3TR, DFL, SPP, SPPF, Bottleneck, BottleneckCSP, C2f, C3Ghost, C3x, GhostBottleneck,
-                    HGBlock, HGStem, Proto, RepC3)
-from .conv import (CBAM, ChannelAttention, Concat, Conv, Conv2, ConvTranspose, DWConv, DWConvTranspose2d, Focus,
-                   GhostConv, LightConv, RepConv, SpatialAttention)
-from .head import Classify, Detect, Pose, RTDETRDecoder, Segment
-from .transformer import (AIFI, MLP, DeformableTransformerDecoder, DeformableTransformerDecoderLayer, LayerNorm2d,
-                          MLPBlock, MSDeformAttn, TransformerBlock, TransformerEncoderLayer, TransformerLayer)
+from .block import (
+    C1,
+    C2,
+    C3,
+    C3TR,
+    DFL,
+    SPP,
+    SPPF,
+    Bottleneck,
+    BottleneckCSP,
+    C2f,
+    C2fAttn,
+    ImagePoolingAttn,
+    C3Ghost,
+    C3x,
+    GhostBottleneck,
+    HGBlock,
+    HGStem,
+    Proto,
+    RepC3,
+    ResNetLayer,
+    ContrastiveHead,
+    BNContrastiveHead,
+    RepNCSPELAN4,
+    ADown,
+    SPPELAN,
+    CBFuse,
+    CBLinear,
+    Silence,
+    PSA,
+    C2fCIB,
+    SCDown,
+    RepVGGDW
+)
+from .conv import (
+    CBAM,
+    ChannelAttention,
+    Concat,
+    Conv,
+    Conv2,
+    ConvTranspose,
+    DWConv,
+    DWConvTranspose2d,
+    Focus,
+    GhostConv,
+    LightConv,
+    RepConv,
+    SpatialAttention,
+)
+from .head import OBB, Classify, Detect, Pose, RTDETRDecoder, Segment, WorldDetect, v10Detect
+from .transformer import (
+    AIFI,
+    MLP,
+    DeformableTransformerDecoder,
+    DeformableTransformerDecoderLayer,
+    LayerNorm2d,
+    MLPBlock,
+    MSDeformAttn,
+    TransformerBlock,
+    TransformerEncoderLayer,
+    TransformerLayer,
+)
 
-__all__ = ('Conv', 'Conv2', 'LightConv', 'RepConv', 'DWConv', 'DWConvTranspose2d', 'ConvTranspose', 'Focus',
-           'GhostConv', 'ChannelAttention', 'SpatialAttention', 'CBAM', 'Concat', 'TransformerLayer',
-           'TransformerBlock', 'MLPBlock', 'LayerNorm2d', 'DFL', 'HGBlock', 'HGStem', 'SPP', 'SPPF', 'C1', 'C2', 'C3',
-           'C2f', 'C3x', 'C3TR', 'C3Ghost', 'GhostBottleneck', 'Bottleneck', 'BottleneckCSP', 'Proto', 'Detect',
-           'Segment', 'Pose', 'Classify', 'TransformerEncoderLayer', 'RepC3', 'RTDETRDecoder', 'AIFI',
-           'DeformableTransformerDecoder', 'DeformableTransformerDecoderLayer', 'MSDeformAttn', 'MLP')
+__all__ = (
+    "Conv",
+    "Conv2",
+    "LightConv",
+    "RepConv",
+    "DWConv",
+    "DWConvTranspose2d",
+    "ConvTranspose",
+    "Focus",
+    "GhostConv",
+    "ChannelAttention",
+    "SpatialAttention",
+    "CBAM",
+    "Concat",
+    "TransformerLayer",
+    "TransformerBlock",
+    "MLPBlock",
+    "LayerNorm2d",
+    "DFL",
+    "HGBlock",
+    "HGStem",
+    "SPP",
+    "SPPF",
+    "C1",
+    "C2",
+    "C3",
+    "C2f",
+    "C2fAttn",
+    "C3x",
+    "C3TR",
+    "C3Ghost",
+    "GhostBottleneck",
+    "Bottleneck",
+    "BottleneckCSP",
+    "Proto",
+    "Detect",
+    "Segment",
+    "Pose",
+    "Classify",
+    "TransformerEncoderLayer",
+    "RepC3",
+    "RTDETRDecoder",
+    "AIFI",
+    "DeformableTransformerDecoder",
+    "DeformableTransformerDecoderLayer",
+    "MSDeformAttn",
+    "MLP",
+    "ResNetLayer",
+    "OBB",
+    "WorldDetect",
+    "ImagePoolingAttn",
+    "ContrastiveHead",
+    "BNContrastiveHead",
+    "RepNCSPELAN4",
+    "ADown",
+    "SPPELAN",
+    "CBFuse",
+    "CBLinear",
+    "Silence",
+    "PSA",
+    "C2fCIB",
+    "SCDown",
+    "RepVGGDW",
+    "v10Detect"
+)

ultralytics/nn/modules/block.py

@@ -1,22 +1,50 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
-"""
-Block modules
-"""
+"""Block modules."""
 
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
-from .conv import Conv, DWConv, GhostConv, LightConv, RepConv
+from .conv import Conv, DWConv, GhostConv, LightConv, RepConv, autopad
 from .transformer import TransformerBlock
+from ultralytics.utils.torch_utils import fuse_conv_and_bn
 
-__all__ = ('DFL', 'HGBlock', 'HGStem', 'SPP', 'SPPF', 'C1', 'C2', 'C3', 'C2f', 'C3x', 'C3TR', 'C3Ghost',
-           'GhostBottleneck', 'Bottleneck', 'BottleneckCSP', 'Proto', 'RepC3')
+__all__ = (
+    "DFL",
+    "HGBlock",
+    "HGStem",
+    "SPP",
+    "SPPF",
+    "C1",
+    "C2",
+    "C3",
+    "C2f",
+    "C2fAttn",
+    "ImagePoolingAttn",
+    "ContrastiveHead",
+    "BNContrastiveHead",
+    "C3x",
+    "C3TR",
+    "C3Ghost",
+    "GhostBottleneck",
+    "Bottleneck",
+    "BottleneckCSP",
+    "Proto",
+    "RepC3",
+    "ResNetLayer",
+    "RepNCSPELAN4",
+    "ADown",
+    "SPPELAN",
+    "CBFuse",
+    "CBLinear",
+    "Silence",
+)
 
 
 class DFL(nn.Module):
     """
     Integral module of Distribution Focal Loss (DFL).
+
     Proposed in Generalized Focal Loss https://ieeexplore.ieee.org/document/9792391
     """
 
@@ -30,7 +58,7 @@ class DFL(nn.Module):
 
     def forward(self, x):
         """Applies a transformer layer on input tensor 'x' and returns a tensor."""
-        b, c, a = x.shape  # batch, channels, anchors
+        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)
         # return self.conv(x.view(b, self.c1, 4, a).softmax(1)).view(b, 4, a)
 
@@ -38,7 +66,12 @@ class DFL(nn.Module):
 class Proto(nn.Module):
     """YOLOv8 mask Proto module for segmentation models."""
 
-    def __init__(self, c1, c_=256, c2=32):  # ch_in, number of protos, number of masks
+    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')
@@ -51,11 +84,14 @@ class Proto(nn.Module):
 
 
 class HGStem(nn.Module):
-    """StemBlock of PPHGNetV2 with 5 convolutions and one maxpool2d.
+    """
+    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())
@@ -79,11 +115,14 @@ class HGStem(nn.Module):
 
 
 class HGBlock(nn.Module):
-    """HG_Block of PPHGNetV2 with 2 convolutions and LightConv.
+    """
+    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))
@@ -119,7 +158,12 @@ class SPP(nn.Module):
 class SPPF(nn.Module):
     """Spatial Pyramid Pooling - Fast (SPPF) layer for YOLOv5 by Glenn Jocher."""
 
-    def __init__(self, c1, c2, k=5):  # equivalent to SPP(k=(5, 9, 13))
+    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)
@@ -137,7 +181,8 @@ class SPPF(nn.Module):
 class C1(nn.Module):
     """CSP Bottleneck with 1 convolution."""
 
-    def __init__(self, c1, c2, n=1):  # ch_in, ch_out, number
+    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)))
@@ -151,7 +196,10 @@ class C1(nn.Module):
 class C2(nn.Module):
     """CSP Bottleneck with 2 convolutions."""
 
-    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
+    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)
@@ -168,7 +216,10 @@ class C2(nn.Module):
 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):  # ch_in, ch_out, number, shortcut, groups, expansion
+    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)
@@ -191,7 +242,8 @@ class C2f(nn.Module):
 class C3(nn.Module):
     """CSP Bottleneck with 3 convolutions."""
 
-    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
+    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)
@@ -218,6 +270,7 @@ 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)
@@ -253,15 +306,18 @@ class C3Ghost(C3):
 class GhostBottleneck(nn.Module):
     """Ghost Bottleneck https://github.com/huawei-noah/ghostnet."""
 
-    def __init__(self, c1, c2, k=3, s=1):  # ch_in, ch_out, kernel, stride
+    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()
+            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."""
@@ -271,7 +327,10 @@ class GhostBottleneck(nn.Module):
 class Bottleneck(nn.Module):
     """Standard bottleneck."""
 
-    def __init__(self, c1, c2, shortcut=True, g=1, k=(3, 3), e=0.5):  # ch_in, ch_out, shortcut, groups, kernels, expand
+    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)
@@ -279,14 +338,15 @@ class Bottleneck(nn.Module):
         self.add = shortcut and c1 == c2
 
     def forward(self, x):
-        """'forward()' applies the YOLOv5 FPN to input data."""
+        """'forward()' applies the YOLO FPN to input data."""
         return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
 
 
 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):  # ch_in, ch_out, number, shortcut, groups, expansion
+    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)
@@ -302,3 +362,466 @@ class BottleneckCSP(nn.Module):
         y1 = self.cv3(self.m(self.cv1(x)))
         y2 = self.cv2(x)
         return self.cv4(self.act(self.bn(torch.cat((y1, y2), 1))))
+
+
+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))
+
+
+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)
+
+
+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)
+
+
+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))
+
+
+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
+
+
+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__()
+        self.bias = nn.Parameter(torch.zeros([]))
+        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
+
+
+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)
+        self.bias = nn.Parameter(torch.zeros([]))
+        # 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
+
+
+class RepBottleneck(nn.Module):
+    """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__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = RepConv(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 pass through RepBottleneck layer."""
+        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
+
+
+class RepCSP(nn.Module):
+    """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__()
+        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(*(RepBottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))
+
+    def forward(self, x):
+        """Forward pass through RepCSP layer."""
+        return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(x)), 1))
+
+
+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))
+
+
+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)
+
+
+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))
+
+
+class Silence(nn.Module):
+    """Silence."""
+
+    def __init__(self):
+        """Initializes the Silence module."""
+        super(Silence, self).__init__()
+
+    def forward(self, x):
+        """Forward pass through Silence layer."""
+        return x
+
+
+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."""
+        outs = self.conv(x).split(self.c2s, dim=1)
+        return outs
+
+
+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])]
+        out = torch.sum(torch.stack(res + xs[-1:]), dim=0)
+        return out
+
+
+class RepVGGDW(torch.nn.Module):
+    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):
+        return self.act(self.conv(x) + self.conv1(x))
+    
+    def forward_fuse(self, x):
+        return self.act(self.conv(x))
+
+    @torch.no_grad()
+    def fuse(self):
+        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
+
+class CIB(nn.Module):
+    """Standard bottleneck."""
+
+    def __init__(self, c1, c2, shortcut=True, e=0.5, lk=False):
+        """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 = nn.Sequential(
+            Conv(c1, c1, 3, g=c1),
+            Conv(c1, 2 * c_, 1),
+            Conv(2 * c_, 2 * c_, 3, g=2 * c_) if not lk else RepVGGDW(2 * c_),
+            Conv(2 * c_, c2, 1),
+            Conv(c2, c2, 3, g=c2),
+        )
+
+        self.add = shortcut and c1 == c2
+
+    def forward(self, x):
+        """'forward()' applies the YOLO FPN to input data."""
+        return x + self.cv1(x) if self.add else self.cv1(x)
+
+class C2fCIB(C2f):
+    """Faster Implementation of CSP Bottleneck with 2 convolutions."""
+
+    def __init__(self, c1, c2, n=1, shortcut=False, lk=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__(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))
+
+
+class Attention(nn.Module):
+    def __init__(self, dim, num_heads=8,
+                 attn_ratio=0.5):
+        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):
+        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
+
+class PSA(nn.Module):
+
+    def __init__(self, c1, c2, e=0.5):
+        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):
+        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))
+
+class SCDown(nn.Module):
+    def __init__(self, c1, c2, k, s):
+        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):
+        return self.cv2(self.cv1(x))

ultralytics/nn/modules/conv.py

@@ -1,7 +1,5 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
-"""
-Convolution modules
-"""
+"""Convolution modules."""
 
 import math
 
@@ -9,8 +7,21 @@ import numpy as np
 import torch
 import torch.nn as nn
 
-__all__ = ('Conv', 'Conv2', 'LightConv', 'DWConv', 'DWConvTranspose2d', 'ConvTranspose', 'Focus', 'GhostConv',
-           'ChannelAttention', 'SpatialAttention', 'CBAM', 'Concat', 'RepConv')
+__all__ = (
+    "Conv",
+    "Conv2",
+    "LightConv",
+    "DWConv",
+    "DWConvTranspose2d",
+    "ConvTranspose",
+    "Focus",
+    "GhostConv",
+    "ChannelAttention",
+    "SpatialAttention",
+    "CBAM",
+    "Concat",
+    "RepConv",
+)
 
 
 def autopad(k, p=None, d=1):  # kernel, padding, dilation
@@ -24,6 +35,7 @@ def autopad(k, p=None, d=1):  # kernel, padding, dilation
 
 class Conv(nn.Module):
     """Standard convolution with args(ch_in, ch_out, kernel, stride, padding, groups, dilation, activation)."""
+
     default_act = nn.SiLU()  # default activation
 
     def __init__(self, c1, c2, k=1, s=1, p=None, g=1, d=1, act=True):
@@ -62,14 +74,16 @@ class Conv2(Conv):
         """Fuse parallel convolutions."""
         w = torch.zeros_like(self.conv.weight.data)
         i = [x // 2 for x in w.shape[2:]]
-        w[:, :, i[0]:i[0] + 1, i[1]:i[1] + 1] = self.cv2.weight.data.clone()
+        w[:, :, i[0] : i[0] + 1, i[1] : i[1] + 1] = self.cv2.weight.data.clone()
         self.conv.weight.data += w
-        self.__delattr__('cv2')
+        self.__delattr__("cv2")
         self.forward = self.forward_fuse
 
 
 class LightConv(nn.Module):
-    """Light convolution with args(ch_in, ch_out, kernel).
+    """
+    Light convolution with args(ch_in, ch_out, kernel).
+
     https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/backbones/hgnet_v2.py
     """
 
@@ -88,6 +102,7 @@ class DWConv(Conv):
     """Depth-wise convolution."""
 
     def __init__(self, c1, c2, k=1, s=1, d=1, act=True):  # ch_in, ch_out, kernel, stride, dilation, activation
+        """Initialize Depth-wise convolution with given parameters."""
         super().__init__(c1, c2, k, s, g=math.gcd(c1, c2), d=d, act=act)
 
 
@@ -95,11 +110,13 @@ class DWConvTranspose2d(nn.ConvTranspose2d):
     """Depth-wise transpose convolution."""
 
     def __init__(self, c1, c2, k=1, s=1, p1=0, p2=0):  # ch_in, ch_out, kernel, stride, padding, padding_out
+        """Initialize DWConvTranspose2d class with given parameters."""
         super().__init__(c1, c2, k, s, p1, p2, groups=math.gcd(c1, c2))
 
 
 class ConvTranspose(nn.Module):
     """Convolution transpose 2d layer."""
+
     default_act = nn.SiLU()  # default activation
 
     def __init__(self, c1, c2, k=2, s=2, p=0, bn=True, act=True):
@@ -121,12 +138,18 @@ class ConvTranspose(nn.Module):
 class Focus(nn.Module):
     """Focus wh information into c-space."""
 
-    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):
+        """Initializes Focus object with user defined channel, convolution, padding, group and activation values."""
         super().__init__()
         self.conv = Conv(c1 * 4, c2, k, s, p, g, act=act)
         # self.contract = Contract(gain=2)
 
-    def forward(self, x):  # x(b,c,w,h) -> y(b,4c,w/2,h/2)
+    def forward(self, x):
+        """
+        Applies convolution to concatenated tensor and returns the output.
+
+        Input shape is (b,c,w,h) and output shape is (b,4c,w/2,h/2).
+        """
         return self.conv(torch.cat((x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]), 1))
         # return self.conv(self.contract(x))
 
@@ -134,7 +157,10 @@ class Focus(nn.Module):
 class GhostConv(nn.Module):
     """Ghost Convolution https://github.com/huawei-noah/ghostnet."""
 
-    def __init__(self, c1, c2, k=1, s=1, g=1, act=True):  # ch_in, ch_out, kernel, stride, groups
+    def __init__(self, c1, c2, k=1, s=1, g=1, act=True):
+        """Initializes the GhostConv object with input channels, output channels, kernel size, stride, groups and
+        activation.
+        """
         super().__init__()
         c_ = c2 // 2  # hidden channels
         self.cv1 = Conv(c1, c_, k, s, None, g, act=act)
@@ -148,12 +174,16 @@ class GhostConv(nn.Module):
 
 class RepConv(nn.Module):
     """
-    RepConv is a basic rep-style block, including training and deploy status. This module is used in RT-DETR.
+    RepConv is a basic rep-style block, including training and deploy status.
+
+    This module is used in RT-DETR.
     Based on https://github.com/DingXiaoH/RepVGG/blob/main/repvgg.py
     """
+
     default_act = nn.SiLU()  # default activation
 
     def __init__(self, c1, c2, k=3, s=1, p=1, g=1, d=1, act=True, bn=False, deploy=False):
+        """Initializes Light Convolution layer with inputs, outputs & optional activation function."""
         super().__init__()
         assert k == 3 and p == 1
         self.g = g
@@ -166,27 +196,30 @@ class RepConv(nn.Module):
         self.conv2 = Conv(c1, c2, 1, s, p=(p - k // 2), g=g, act=False)
 
     def forward_fuse(self, x):
-        """Forward process"""
+        """Forward process."""
         return self.act(self.conv(x))
 
     def forward(self, x):
-        """Forward process"""
+        """Forward process."""
         id_out = 0 if self.bn is None else self.bn(x)
         return self.act(self.conv1(x) + self.conv2(x) + id_out)
 
     def get_equivalent_kernel_bias(self):
+        """Returns equivalent kernel and bias by adding 3x3 kernel, 1x1 kernel and identity kernel with their biases."""
         kernel3x3, bias3x3 = self._fuse_bn_tensor(self.conv1)
         kernel1x1, bias1x1 = self._fuse_bn_tensor(self.conv2)
         kernelid, biasid = self._fuse_bn_tensor(self.bn)
         return kernel3x3 + self._pad_1x1_to_3x3_tensor(kernel1x1) + kernelid, bias3x3 + bias1x1 + biasid
 
     def _pad_1x1_to_3x3_tensor(self, kernel1x1):
+        """Pads a 1x1 tensor to a 3x3 tensor."""
         if kernel1x1 is None:
             return 0
         else:
             return torch.nn.functional.pad(kernel1x1, [1, 1, 1, 1])
 
     def _fuse_bn_tensor(self, branch):
+        """Generates appropriate kernels and biases for convolution by fusing branches of the neural network."""
         if branch is None:
             return 0, 0
         if isinstance(branch, Conv):
@@ -197,7 +230,7 @@ class RepConv(nn.Module):
             beta = branch.bn.bias
             eps = branch.bn.eps
         elif isinstance(branch, nn.BatchNorm2d):
-            if not hasattr(self, 'id_tensor'):
+            if not hasattr(self, "id_tensor"):
                 input_dim = self.c1 // self.g
                 kernel_value = np.zeros((self.c1, input_dim, 3, 3), dtype=np.float32)
                 for i in range(self.c1):
@@ -214,41 +247,46 @@ class RepConv(nn.Module):
         return kernel * t, beta - running_mean * gamma / std
 
     def fuse_convs(self):
-        if hasattr(self, 'conv'):
+        """Combines two convolution layers into a single layer and removes unused attributes from the class."""
+        if hasattr(self, "conv"):
             return
         kernel, bias = self.get_equivalent_kernel_bias()
-        self.conv = nn.Conv2d(in_channels=self.conv1.conv.in_channels,
-                              out_channels=self.conv1.conv.out_channels,
-                              kernel_size=self.conv1.conv.kernel_size,
-                              stride=self.conv1.conv.stride,
-                              padding=self.conv1.conv.padding,
-                              dilation=self.conv1.conv.dilation,
-                              groups=self.conv1.conv.groups,
-                              bias=True).requires_grad_(False)
+        self.conv = nn.Conv2d(
+            in_channels=self.conv1.conv.in_channels,
+            out_channels=self.conv1.conv.out_channels,
+            kernel_size=self.conv1.conv.kernel_size,
+            stride=self.conv1.conv.stride,
+            padding=self.conv1.conv.padding,
+            dilation=self.conv1.conv.dilation,
+            groups=self.conv1.conv.groups,
+            bias=True,
+        ).requires_grad_(False)
         self.conv.weight.data = kernel
         self.conv.bias.data = bias
         for para in self.parameters():
             para.detach_()
-        self.__delattr__('conv1')
-        self.__delattr__('conv2')
-        if hasattr(self, 'nm'):
-            self.__delattr__('nm')
-        if hasattr(self, 'bn'):
-            self.__delattr__('bn')
-        if hasattr(self, 'id_tensor'):
-            self.__delattr__('id_tensor')
+        self.__delattr__("conv1")
+        self.__delattr__("conv2")
+        if hasattr(self, "nm"):
+            self.__delattr__("nm")
+        if hasattr(self, "bn"):
+            self.__delattr__("bn")
+        if hasattr(self, "id_tensor"):
+            self.__delattr__("id_tensor")
 
 
 class ChannelAttention(nn.Module):
     """Channel-attention module https://github.com/open-mmlab/mmdetection/tree/v3.0.0rc1/configs/rtmdet."""
 
     def __init__(self, channels: int) -> None:
+        """Initializes the class and sets the basic configurations and instance variables required."""
         super().__init__()
         self.pool = nn.AdaptiveAvgPool2d(1)
         self.fc = nn.Conv2d(channels, channels, 1, 1, 0, bias=True)
         self.act = nn.Sigmoid()
 
     def forward(self, x: torch.Tensor) -> torch.Tensor:
+        """Applies forward pass using activation on convolutions of the input, optionally using batch normalization."""
         return x * self.act(self.fc(self.pool(x)))
 
 
@@ -258,7 +296,7 @@ class SpatialAttention(nn.Module):
     def __init__(self, kernel_size=7):
         """Initialize Spatial-attention module with kernel size argument."""
         super().__init__()
-        assert kernel_size in (3, 7), 'kernel size must be 3 or 7'
+        assert kernel_size in (3, 7), "kernel size must be 3 or 7"
         padding = 3 if kernel_size == 7 else 1
         self.cv1 = nn.Conv2d(2, 1, kernel_size, padding=padding, bias=False)
         self.act = nn.Sigmoid()
@@ -271,7 +309,8 @@ class SpatialAttention(nn.Module):
 class CBAM(nn.Module):
     """Convolutional Block Attention Module."""
 
-    def __init__(self, c1, kernel_size=7):  # ch_in, kernels
+    def __init__(self, c1, kernel_size=7):
+        """Initialize CBAM with given input channel (c1) and kernel size."""
         super().__init__()
         self.channel_attention = ChannelAttention(c1)
         self.spatial_attention = SpatialAttention(kernel_size)

ultralytics/nn/modules/head.py

@@ -1,7 +1,5 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
-"""
-Model head modules
-"""
+"""Model head modules."""
 
 import math
 
@@ -9,25 +7,28 @@ import torch
 import torch.nn as nn
 from torch.nn.init import constant_, xavier_uniform_
 
-from ultralytics.utils.tal import TORCH_1_10, dist2bbox, make_anchors
-
-from .block import DFL, Proto
+from ultralytics.utils.tal import TORCH_1_10, dist2bbox, dist2rbox, make_anchors
+from .block import DFL, Proto, ContrastiveHead, BNContrastiveHead
 from .conv import Conv
 from .transformer import MLP, DeformableTransformerDecoder, DeformableTransformerDecoderLayer
-from .utils import bias_init_with_prob, linear_init_
+from .utils import bias_init_with_prob, linear_init
+import copy
+from ultralytics.utils import ops
 
-__all__ = 'Detect', 'Segment', 'Pose', 'Classify', 'RTDETRDecoder'
+__all__ = "Detect", "Segment", "Pose", "Classify", "OBB", "RTDETRDecoder"
 
 
 class Detect(nn.Module):
     """YOLOv8 Detect head for detection models."""
+
     dynamic = False  # force grid reconstruction
     export = False  # export mode
     shape = None
     anchors = torch.empty(0)  # init
     strides = torch.empty(0)  # init
 
-    def __init__(self, nc=80, ch=()):  # detection layer
+    def __init__(self, nc=80, ch=()):
+        """Initializes the YOLOv8 detection layer with specified number of classes and channels."""
         super().__init__()
         self.nc = nc  # number of classes
         self.nl = len(ch)  # number of detection layers
@@ -36,41 +37,54 @@ class Detect(nn.Module):
         self.stride = torch.zeros(self.nl)  # strides computed during build
         c2, c3 = max((16, ch[0] // 4, self.reg_max * 4)), max(ch[0], min(self.nc, 100))  # channels
         self.cv2 = nn.ModuleList(
-            nn.Sequential(Conv(x, c2, 3), Conv(c2, c2, 3), nn.Conv2d(c2, 4 * self.reg_max, 1)) for x in ch)
+            nn.Sequential(Conv(x, c2, 3), Conv(c2, c2, 3), nn.Conv2d(c2, 4 * self.reg_max, 1)) for x in ch
+        )
         self.cv3 = nn.ModuleList(nn.Sequential(Conv(x, c3, 3), Conv(c3, c3, 3), nn.Conv2d(c3, self.nc, 1)) for x in ch)
         self.dfl = DFL(self.reg_max) if self.reg_max > 1 else nn.Identity()
 
-    def forward(self, x):
-        """Concatenates and returns predicted bounding boxes and class probabilities."""
+    def inference(self, x):
+        # Inference path
         shape = x[0].shape  # BCHW
-        for i in range(self.nl):
-            x[i] = torch.cat((self.cv2[i](x[i]), self.cv3[i](x[i])), 1)
-        if self.training:
-            return x
-        elif self.dynamic or self.shape != shape:
+        x_cat = torch.cat([xi.view(shape[0], self.no, -1) for xi in x], 2)
+        if self.dynamic or self.shape != shape:
             self.anchors, self.strides = (x.transpose(0, 1) for x in make_anchors(x, self.stride, 0.5))
             self.shape = shape
 
-        x_cat = torch.cat([xi.view(shape[0], self.no, -1) for xi in x], 2)
-        if self.export and self.format in ('saved_model', 'pb', 'tflite', 'edgetpu', 'tfjs'):  # avoid TF FlexSplitV ops
-            box = x_cat[:, :self.reg_max * 4]
-            cls = x_cat[:, self.reg_max * 4:]
+        if self.export and self.format in ("saved_model", "pb", "tflite", "edgetpu", "tfjs"):  # avoid TF FlexSplitV ops
+            box = x_cat[:, : self.reg_max * 4]
+            cls = x_cat[:, self.reg_max * 4 :]
         else:
             box, cls = x_cat.split((self.reg_max * 4, self.nc), 1)
-        dbox = dist2bbox(self.dfl(box), self.anchors.unsqueeze(0), xywh=True, dim=1) * self.strides
 
-        if self.export and self.format in ('tflite', 'edgetpu'):
-            # Normalize xywh with image size to mitigate quantization error of TFLite integer models as done in YOLOv5:
-            # https://github.com/ultralytics/yolov5/blob/0c8de3fca4a702f8ff5c435e67f378d1fce70243/models/tf.py#L307-L309
-            # See this PR for details: https://github.com/ultralytics/ultralytics/pull/1695
-            img_h = shape[2] * self.stride[0]
-            img_w = shape[3] * self.stride[0]
-            img_size = torch.tensor([img_w, img_h, img_w, img_h], device=dbox.device).reshape(1, 4, 1)
-            dbox /= img_size
+        if self.export and self.format in ("tflite", "edgetpu"):
+            # Precompute normalization factor to increase numerical stability
+            # See https://github.com/ultralytics/ultralytics/issues/7371
+            grid_h = shape[2]
+            grid_w = shape[3]
+            grid_size = torch.tensor([grid_w, grid_h, grid_w, grid_h], device=box.device).reshape(1, 4, 1)
+            norm = self.strides / (self.stride[0] * grid_size)
+            dbox = self.decode_bboxes(self.dfl(box) * norm, self.anchors.unsqueeze(0) * norm[:, :2])
+        else:
+            dbox = self.decode_bboxes(self.dfl(box), self.anchors.unsqueeze(0)) * self.strides
 
         y = torch.cat((dbox, cls.sigmoid()), 1)
         return y if self.export else (y, x)
 
+    def forward_feat(self, x, cv2, cv3):
+        y = []
+        for i in range(self.nl):
+            y.append(torch.cat((cv2[i](x[i]), cv3[i](x[i])), 1))
+        return y
+
+    def forward(self, x):
+        """Concatenates and returns predicted bounding boxes and class probabilities."""
+        y = self.forward_feat(x, self.cv2, self.cv3)
+        
+        if self.training:
+            return y
+
+        return self.inference(y)
+
     def bias_init(self):
         """Initialize Detect() biases, WARNING: requires stride availability."""
         m = self  # self.model[-1]  # Detect() module
@@ -78,7 +92,13 @@ class Detect(nn.Module):
         # ncf = math.log(0.6 / (m.nc - 0.999999)) if cf is None else torch.log(cf / cf.sum())  # nominal class frequency
         for a, b, s in zip(m.cv2, m.cv3, m.stride):  # from
             a[-1].bias.data[:] = 1.0  # box
-            b[-1].bias.data[:m.nc] = math.log(5 / m.nc / (640 / s) ** 2)  # cls (.01 objects, 80 classes, 640 img)
+            b[-1].bias.data[: m.nc] = math.log(5 / m.nc / (640 / s) ** 2)  # cls (.01 objects, 80 classes, 640 img)
+
+    def decode_bboxes(self, bboxes, anchors):
+        """Decode bounding boxes."""
+        if self.export:
+            return dist2bbox(bboxes, anchors, xywh=False, dim=1)
+        return dist2bbox(bboxes, anchors, xywh=True, dim=1)
 
 
 class Segment(Detect):
@@ -107,6 +127,37 @@ class Segment(Detect):
         return (torch.cat([x, mc], 1), p) if self.export else (torch.cat([x[0], mc], 1), (x[1], mc, p))
 
 
+class OBB(Detect):
+    """YOLOv8 OBB detection head for detection with rotation models."""
+
+    def __init__(self, nc=80, ne=1, ch=()):
+        """Initialize OBB with number of classes `nc` and layer channels `ch`."""
+        super().__init__(nc, ch)
+        self.ne = ne  # number of extra parameters
+        self.detect = Detect.forward
+
+        c4 = max(ch[0] // 4, self.ne)
+        self.cv4 = nn.ModuleList(nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3), nn.Conv2d(c4, self.ne, 1)) for x in ch)
+
+    def forward(self, x):
+        """Concatenates and returns predicted bounding boxes and class probabilities."""
+        bs = x[0].shape[0]  # batch size
+        angle = torch.cat([self.cv4[i](x[i]).view(bs, self.ne, -1) for i in range(self.nl)], 2)  # OBB theta logits
+        # NOTE: set `angle` as an attribute so that `decode_bboxes` could use it.
+        angle = (angle.sigmoid() - 0.25) * math.pi  # [-pi/4, 3pi/4]
+        # angle = angle.sigmoid() * math.pi / 2  # [0, pi/2]
+        if not self.training:
+            self.angle = angle
+        x = self.detect(self, x)
+        if self.training:
+            return x, angle
+        return torch.cat([x, angle], 1) if self.export else (torch.cat([x[0], angle], 1), (x[1], angle))
+
+    def decode_bboxes(self, bboxes, anchors):
+        """Decode rotated bounding boxes."""
+        return dist2rbox(bboxes, self.angle, anchors, dim=1)
+
+
 class Pose(Detect):
     """YOLOv8 Pose head for keypoints models."""
 
@@ -142,7 +193,7 @@ class Pose(Detect):
         else:
             y = kpts.clone()
             if ndim == 3:
-                y[:, 2::3].sigmoid_()  # inplace sigmoid
+                y[:, 2::3] = y[:, 2::3].sigmoid()  # sigmoid (WARNING: inplace .sigmoid_() Apple MPS bug)
             y[:, 0::ndim] = (y[:, 0::ndim] * 2.0 + (self.anchors[0] - 0.5)) * self.strides
             y[:, 1::ndim] = (y[:, 1::ndim] * 2.0 + (self.anchors[1] - 0.5)) * self.strides
             return y
@@ -151,7 +202,10 @@ class Pose(Detect):
 class Classify(nn.Module):
     """YOLOv8 classification head, i.e. x(b,c1,20,20) to x(b,c2)."""
 
-    def __init__(self, c1, c2, k=1, s=1, p=None, g=1):  # ch_in, ch_out, kernel, stride, padding, groups
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1):
+        """Initializes YOLOv8 classification head with specified input and output channels, kernel size, stride,
+        padding, and groups.
+        """
         super().__init__()
         c_ = 1280  # efficientnet_b0 size
         self.conv = Conv(c1, c_, k, s, p, g)
@@ -167,27 +221,99 @@ class Classify(nn.Module):
         return x if self.training else x.softmax(1)
 
 
+class WorldDetect(Detect):
+    def __init__(self, nc=80, embed=512, with_bn=False, ch=()):
+        """Initialize YOLOv8 detection layer with nc classes and layer channels ch."""
+        super().__init__(nc, ch)
+        c3 = max(ch[0], min(self.nc, 100))
+        self.cv3 = nn.ModuleList(nn.Sequential(Conv(x, c3, 3), Conv(c3, c3, 3), nn.Conv2d(c3, embed, 1)) for x in ch)
+        self.cv4 = nn.ModuleList(BNContrastiveHead(embed) if with_bn else ContrastiveHead() for _ in ch)
+
+    def forward(self, x, text):
+        """Concatenates and returns predicted bounding boxes and class probabilities."""
+        for i in range(self.nl):
+            x[i] = torch.cat((self.cv2[i](x[i]), self.cv4[i](self.cv3[i](x[i]), text)), 1)
+        if self.training:
+            return x
+
+        # Inference path
+        shape = x[0].shape  # BCHW
+        x_cat = torch.cat([xi.view(shape[0], self.nc + self.reg_max * 4, -1) for xi in x], 2)
+        if self.dynamic or self.shape != shape:
+            self.anchors, self.strides = (x.transpose(0, 1) for x in make_anchors(x, self.stride, 0.5))
+            self.shape = shape
+
+        if self.export and self.format in ("saved_model", "pb", "tflite", "edgetpu", "tfjs"):  # avoid TF FlexSplitV ops
+            box = x_cat[:, : self.reg_max * 4]
+            cls = x_cat[:, self.reg_max * 4 :]
+        else:
+            box, cls = x_cat.split((self.reg_max * 4, self.nc), 1)
+
+        if self.export and self.format in ("tflite", "edgetpu"):
+            # Precompute normalization factor to increase numerical stability
+            # See https://github.com/ultralytics/ultralytics/issues/7371
+            grid_h = shape[2]
+            grid_w = shape[3]
+            grid_size = torch.tensor([grid_w, grid_h, grid_w, grid_h], device=box.device).reshape(1, 4, 1)
+            norm = self.strides / (self.stride[0] * grid_size)
+            dbox = self.decode_bboxes(self.dfl(box) * norm, self.anchors.unsqueeze(0) * norm[:, :2])
+        else:
+            dbox = self.decode_bboxes(self.dfl(box), self.anchors.unsqueeze(0)) * self.strides
+
+        y = torch.cat((dbox, cls.sigmoid()), 1)
+        return y if self.export else (y, x)
+
+
 class RTDETRDecoder(nn.Module):
+    """
+    Real-Time Deformable Transformer Decoder (RTDETRDecoder) module for object detection.
+
+    This decoder module utilizes Transformer architecture along with deformable convolutions to predict bounding boxes
+    and class labels for objects in an image. It integrates features from multiple layers and runs through a series of
+    Transformer decoder layers to output the final predictions.
+    """
+
     export = False  # export mode
 
     def __init__(
-            self,
-            nc=80,
-            ch=(512, 1024, 2048),
-            hd=256,  # hidden dim
-            nq=300,  # num queries
-            ndp=4,  # num decoder points
-            nh=8,  # num head
-            ndl=6,  # num decoder layers
-            d_ffn=1024,  # dim of feedforward
-            dropout=0.,
-            act=nn.ReLU(),
-            eval_idx=-1,
-            # training args
-            nd=100,  # num denoising
-            label_noise_ratio=0.5,
-            box_noise_scale=1.0,
-            learnt_init_query=False):
+        self,
+        nc=80,
+        ch=(512, 1024, 2048),
+        hd=256,  # hidden dim
+        nq=300,  # num queries
+        ndp=4,  # num decoder points
+        nh=8,  # num head
+        ndl=6,  # num decoder layers
+        d_ffn=1024,  # dim of feedforward
+        dropout=0.0,
+        act=nn.ReLU(),
+        eval_idx=-1,
+        # Training args
+        nd=100,  # num denoising
+        label_noise_ratio=0.5,
+        box_noise_scale=1.0,
+        learnt_init_query=False,
+    ):
+        """
+        Initializes the RTDETRDecoder module with the given parameters.
+
+        Args:
+            nc (int): Number of classes. Default is 80.
+            ch (tuple): Channels in the backbone feature maps. Default is (512, 1024, 2048).
+            hd (int): Dimension of hidden layers. Default is 256.
+            nq (int): Number of query points. Default is 300.
+            ndp (int): Number of decoder points. Default is 4.
+            nh (int): Number of heads in multi-head attention. Default is 8.
+            ndl (int): Number of decoder layers. Default is 6.
+            d_ffn (int): Dimension of the feed-forward networks. Default is 1024.
+            dropout (float): Dropout rate. Default is 0.
+            act (nn.Module): Activation function. Default is nn.ReLU.
+            eval_idx (int): Evaluation index. Default is -1.
+            nd (int): Number of denoising. Default is 100.
+            label_noise_ratio (float): Label noise ratio. Default is 0.5.
+            box_noise_scale (float): Box noise scale. Default is 1.0.
+            learnt_init_query (bool): Whether to learn initial query embeddings. Default is False.
+        """
         super().__init__()
         self.hidden_dim = hd
         self.nhead = nh
@@ -196,7 +322,7 @@ class RTDETRDecoder(nn.Module):
         self.num_queries = nq
         self.num_decoder_layers = ndl
 
-        # backbone feature projection
+        # Backbone feature projection
         self.input_proj = nn.ModuleList(nn.Sequential(nn.Conv2d(x, hd, 1, bias=False), nn.BatchNorm2d(hd)) for x in ch)
         # NOTE: simplified version but it's not consistent with .pt weights.
         # self.input_proj = nn.ModuleList(Conv(x, hd, act=False) for x in ch)
@@ -205,58 +331,61 @@ class RTDETRDecoder(nn.Module):
         decoder_layer = DeformableTransformerDecoderLayer(hd, nh, d_ffn, dropout, act, self.nl, ndp)
         self.decoder = DeformableTransformerDecoder(hd, decoder_layer, ndl, eval_idx)
 
-        # denoising part
+        # Denoising part
         self.denoising_class_embed = nn.Embedding(nc, hd)
         self.num_denoising = nd
         self.label_noise_ratio = label_noise_ratio
         self.box_noise_scale = box_noise_scale
 
-        # decoder embedding
+        # Decoder embedding
         self.learnt_init_query = learnt_init_query
         if learnt_init_query:
             self.tgt_embed = nn.Embedding(nq, hd)
         self.query_pos_head = MLP(4, 2 * hd, hd, num_layers=2)
 
-        # encoder head
+        # Encoder head
         self.enc_output = nn.Sequential(nn.Linear(hd, hd), nn.LayerNorm(hd))
         self.enc_score_head = nn.Linear(hd, nc)
         self.enc_bbox_head = MLP(hd, hd, 4, num_layers=3)
 
-        # decoder head
+        # Decoder head
         self.dec_score_head = nn.ModuleList([nn.Linear(hd, nc) for _ in range(ndl)])
         self.dec_bbox_head = nn.ModuleList([MLP(hd, hd, 4, num_layers=3) for _ in range(ndl)])
 
         self._reset_parameters()
 
     def forward(self, x, batch=None):
+        """Runs the forward pass of the module, returning bounding box and classification scores for the input."""
         from ultralytics.models.utils.ops import get_cdn_group
 
-        # input projection and embedding
+        # Input projection and embedding
         feats, shapes = self._get_encoder_input(x)
 
-        # prepare denoising training
-        dn_embed, dn_bbox, attn_mask, dn_meta = \
-            get_cdn_group(batch,
-                          self.nc,
-                          self.num_queries,
-                          self.denoising_class_embed.weight,
-                          self.num_denoising,
-                          self.label_noise_ratio,
-                          self.box_noise_scale,
-                          self.training)
+        # Prepare denoising training
+        dn_embed, dn_bbox, attn_mask, dn_meta = get_cdn_group(
+            batch,
+            self.nc,
+            self.num_queries,
+            self.denoising_class_embed.weight,
+            self.num_denoising,
+            self.label_noise_ratio,
+            self.box_noise_scale,
+            self.training,
+        )
 
-        embed, refer_bbox, enc_bboxes, enc_scores = \
-            self._get_decoder_input(feats, shapes, dn_embed, dn_bbox)
+        embed, refer_bbox, enc_bboxes, enc_scores = self._get_decoder_input(feats, shapes, dn_embed, dn_bbox)
 
-        # decoder
-        dec_bboxes, dec_scores = self.decoder(embed,
-                                              refer_bbox,
-                                              feats,
-                                              shapes,
-                                              self.dec_bbox_head,
-                                              self.dec_score_head,
-                                              self.query_pos_head,
-                                              attn_mask=attn_mask)
+        # Decoder
+        dec_bboxes, dec_scores = self.decoder(
+            embed,
+            refer_bbox,
+            feats,
+            shapes,
+            self.dec_bbox_head,
+            self.dec_score_head,
+            self.query_pos_head,
+            attn_mask=attn_mask,
+        )
         x = dec_bboxes, dec_scores, enc_bboxes, enc_scores, dn_meta
         if self.training:
             return x
@@ -264,29 +393,31 @@ class RTDETRDecoder(nn.Module):
         y = torch.cat((dec_bboxes.squeeze(0), dec_scores.squeeze(0).sigmoid()), -1)
         return y if self.export else (y, x)
 
-    def _generate_anchors(self, shapes, grid_size=0.05, dtype=torch.float32, device='cpu', eps=1e-2):
+    def _generate_anchors(self, shapes, grid_size=0.05, dtype=torch.float32, device="cpu", eps=1e-2):
+        """Generates anchor bounding boxes for given shapes with specific grid size and validates them."""
         anchors = []
         for i, (h, w) in enumerate(shapes):
             sy = torch.arange(end=h, dtype=dtype, device=device)
             sx = torch.arange(end=w, dtype=dtype, device=device)
-            grid_y, grid_x = torch.meshgrid(sy, sx, indexing='ij') if TORCH_1_10 else torch.meshgrid(sy, sx)
+            grid_y, grid_x = torch.meshgrid(sy, sx, indexing="ij") if TORCH_1_10 else torch.meshgrid(sy, sx)
             grid_xy = torch.stack([grid_x, grid_y], -1)  # (h, w, 2)
 
-            valid_WH = torch.tensor([h, w], dtype=dtype, device=device)
+            valid_WH = torch.tensor([w, h], dtype=dtype, device=device)
             grid_xy = (grid_xy.unsqueeze(0) + 0.5) / valid_WH  # (1, h, w, 2)
-            wh = torch.ones_like(grid_xy, dtype=dtype, device=device) * grid_size * (2.0 ** i)
+            wh = torch.ones_like(grid_xy, dtype=dtype, device=device) * grid_size * (2.0**i)
             anchors.append(torch.cat([grid_xy, wh], -1).view(-1, h * w, 4))  # (1, h*w, 4)
 
         anchors = torch.cat(anchors, 1)  # (1, h*w*nl, 4)
-        valid_mask = ((anchors > eps) * (anchors < 1 - eps)).all(-1, keepdim=True)  # 1, h*w*nl, 1
+        valid_mask = ((anchors > eps) & (anchors < 1 - eps)).all(-1, keepdim=True)  # 1, h*w*nl, 1
         anchors = torch.log(anchors / (1 - anchors))
-        anchors = anchors.masked_fill(~valid_mask, float('inf'))
+        anchors = anchors.masked_fill(~valid_mask, float("inf"))
         return anchors, valid_mask
 
     def _get_encoder_input(self, x):
-        # get projection features
+        """Processes and returns encoder inputs by getting projection features from input and concatenating them."""
+        # Get projection features
         x = [self.input_proj[i](feat) for i, feat in enumerate(x)]
-        # get encoder inputs
+        # Get encoder inputs
         feats = []
         shapes = []
         for feat in x:
@@ -301,14 +432,15 @@ class RTDETRDecoder(nn.Module):
         return feats, shapes
 
     def _get_decoder_input(self, feats, shapes, dn_embed=None, dn_bbox=None):
-        bs = len(feats)
-        # prepare input for decoder
+        """Generates and prepares the input required for the decoder from the provided features and shapes."""
+        bs = feats.shape[0]
+        # Prepare input for decoder
         anchors, valid_mask = self._generate_anchors(shapes, dtype=feats.dtype, device=feats.device)
         features = self.enc_output(valid_mask * feats)  # bs, h*w, 256
 
         enc_outputs_scores = self.enc_score_head(features)  # (bs, h*w, nc)
 
-        # query selection
+        # Query selection
         # (bs, num_queries)
         topk_ind = torch.topk(enc_outputs_scores.max(-1).values, self.num_queries, dim=1).indices.view(-1)
         # (bs, num_queries)
@@ -319,7 +451,7 @@ class RTDETRDecoder(nn.Module):
         # (bs, num_queries, 4)
         top_k_anchors = anchors[:, topk_ind].view(bs, self.num_queries, -1)
 
-        # dynamic anchors + static content
+        # Dynamic anchors + static content
         refer_bbox = self.enc_bbox_head(top_k_features) + top_k_anchors
 
         enc_bboxes = refer_bbox.sigmoid()
@@ -339,20 +471,21 @@ class RTDETRDecoder(nn.Module):
 
     # TODO
     def _reset_parameters(self):
-        # class and bbox head init
+        """Initializes or resets the parameters of the model's various components with predefined weights and biases."""
+        # Class and bbox head init
         bias_cls = bias_init_with_prob(0.01) / 80 * self.nc
-        # NOTE: the weight initialization in `linear_init_` would cause NaN when training with custom datasets.
-        # linear_init_(self.enc_score_head)
+        # NOTE: the weight initialization in `linear_init` would cause NaN when training with custom datasets.
+        # linear_init(self.enc_score_head)
         constant_(self.enc_score_head.bias, bias_cls)
-        constant_(self.enc_bbox_head.layers[-1].weight, 0.)
-        constant_(self.enc_bbox_head.layers[-1].bias, 0.)
+        constant_(self.enc_bbox_head.layers[-1].weight, 0.0)
+        constant_(self.enc_bbox_head.layers[-1].bias, 0.0)
         for cls_, reg_ in zip(self.dec_score_head, self.dec_bbox_head):
-            # linear_init_(cls_)
+            # linear_init(cls_)
             constant_(cls_.bias, bias_cls)
-            constant_(reg_.layers[-1].weight, 0.)
-            constant_(reg_.layers[-1].bias, 0.)
+            constant_(reg_.layers[-1].weight, 0.0)
+            constant_(reg_.layers[-1].bias, 0.0)
 
-        linear_init_(self.enc_output[0])
+        linear_init(self.enc_output[0])
         xavier_uniform_(self.enc_output[0].weight)
         if self.learnt_init_query:
             xavier_uniform_(self.tgt_embed.weight)
@@ -360,3 +493,43 @@ class RTDETRDecoder(nn.Module):
         xavier_uniform_(self.query_pos_head.layers[1].weight)
         for layer in self.input_proj:
             xavier_uniform_(layer[0].weight)
+
+class v10Detect(Detect):
+
+    max_det = 300
+
+    def __init__(self, nc=80, ch=()):
+        super().__init__(nc, ch)
+        c3 = max(ch[0], min(self.nc, 100))  # channels
+        self.cv3 = nn.ModuleList(nn.Sequential(nn.Sequential(Conv(x, x, 3, g=x), Conv(x, c3, 1)), \
+                                               nn.Sequential(Conv(c3, c3, 3, g=c3), Conv(c3, c3, 1)), \
+                                                nn.Conv2d(c3, self.nc, 1)) for i, x in enumerate(ch))
+
+        self.one2one_cv2 = copy.deepcopy(self.cv2)
+        self.one2one_cv3 = copy.deepcopy(self.cv3)
+    
+    def forward(self, x):
+        one2one = self.forward_feat([xi.detach() for xi in x], self.one2one_cv2, self.one2one_cv3)
+        if not self.export:
+            one2many = super().forward(x)
+
+        if not self.training:
+            one2one = self.inference(one2one)
+            if not self.export:
+                return {"one2many": one2many, "one2one": one2one}
+            else:
+                assert(self.max_det != -1)
+                boxes, scores, labels = ops.v10postprocess(one2one.permute(0, 2, 1), self.max_det, self.nc)
+                return torch.cat([boxes, scores.unsqueeze(-1), labels.unsqueeze(-1).to(boxes.dtype)], dim=-1)
+        else:
+            return {"one2many": one2many, "one2one": one2one}
+
+    def bias_init(self):
+        super().bias_init()
+        """Initialize Detect() biases, WARNING: requires stride availability."""
+        m = self  # self.model[-1]  # Detect() module
+        # cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1
+        # ncf = math.log(0.6 / (m.nc - 0.999999)) if cf is None else torch.log(cf / cf.sum())  # nominal class frequency
+        for a, b, s in zip(m.one2one_cv2, m.one2one_cv3, m.stride):  # from
+            a[-1].bias.data[:] = 1.0  # box
+            b[-1].bias.data[: m.nc] = math.log(5 / m.nc / (640 / s) ** 2)  # cls (.01 objects, 80 classes, 640 img)

ultralytics/nn/modules/transformer.py

@@ -1,7 +1,5 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
-"""
-Transformer modules
-"""
+"""Transformer modules."""
 
 import math
 
@@ -13,19 +11,32 @@ from torch.nn.init import constant_, xavier_uniform_
 from .conv import Conv
 from .utils import _get_clones, inverse_sigmoid, multi_scale_deformable_attn_pytorch
 
-__all__ = ('TransformerEncoderLayer', 'TransformerLayer', 'TransformerBlock', 'MLPBlock', 'LayerNorm2d', 'AIFI',
-           'DeformableTransformerDecoder', 'DeformableTransformerDecoderLayer', 'MSDeformAttn', 'MLP')
+__all__ = (
+    "TransformerEncoderLayer",
+    "TransformerLayer",
+    "TransformerBlock",
+    "MLPBlock",
+    "LayerNorm2d",
+    "AIFI",
+    "DeformableTransformerDecoder",
+    "DeformableTransformerDecoderLayer",
+    "MSDeformAttn",
+    "MLP",
+)
 
 
 class TransformerEncoderLayer(nn.Module):
-    """Transformer Encoder."""
+    """Defines a single layer of the transformer encoder."""
 
     def __init__(self, c1, cm=2048, num_heads=8, dropout=0.0, act=nn.GELU(), normalize_before=False):
+        """Initialize the TransformerEncoderLayer with specified parameters."""
         super().__init__()
         from ...utils.torch_utils import TORCH_1_9
+
         if not TORCH_1_9:
             raise ModuleNotFoundError(
-                'TransformerEncoderLayer() requires torch>=1.9 to use nn.MultiheadAttention(batch_first=True).')
+                "TransformerEncoderLayer() requires torch>=1.9 to use nn.MultiheadAttention(batch_first=True)."
+            )
         self.ma = nn.MultiheadAttention(c1, num_heads, dropout=dropout, batch_first=True)
         # Implementation of Feedforward model
         self.fc1 = nn.Linear(c1, cm)
@@ -40,11 +51,13 @@ class TransformerEncoderLayer(nn.Module):
         self.act = act
         self.normalize_before = normalize_before
 
-    def with_pos_embed(self, tensor, pos=None):
-        """Add position embeddings if given."""
+    @staticmethod
+    def with_pos_embed(tensor, pos=None):
+        """Add position embeddings to the tensor if provided."""
         return tensor if pos is None else tensor + pos
 
     def forward_post(self, src, src_mask=None, src_key_padding_mask=None, pos=None):
+        """Performs forward pass with post-normalization."""
         q = k = self.with_pos_embed(src, pos)
         src2 = self.ma(q, k, value=src, attn_mask=src_mask, key_padding_mask=src_key_padding_mask)[0]
         src = src + self.dropout1(src2)
@@ -54,6 +67,7 @@ class TransformerEncoderLayer(nn.Module):
         return self.norm2(src)
 
     def forward_pre(self, src, src_mask=None, src_key_padding_mask=None, pos=None):
+        """Performs forward pass with pre-normalization."""
         src2 = self.norm1(src)
         q = k = self.with_pos_embed(src2, pos)
         src2 = self.ma(q, k, value=src2, attn_mask=src_mask, key_padding_mask=src_key_padding_mask)[0]
@@ -70,27 +84,30 @@ class TransformerEncoderLayer(nn.Module):
 
 
 class AIFI(TransformerEncoderLayer):
+    """Defines the AIFI transformer layer."""
 
     def __init__(self, c1, cm=2048, num_heads=8, dropout=0, act=nn.GELU(), normalize_before=False):
+        """Initialize the AIFI instance with specified parameters."""
         super().__init__(c1, cm, num_heads, dropout, act, normalize_before)
 
     def forward(self, x):
+        """Forward pass for the AIFI transformer layer."""
         c, h, w = x.shape[1:]
         pos_embed = self.build_2d_sincos_position_embedding(w, h, c)
-        # flatten [B, C, H, W] to [B, HxW, C]
+        # Flatten [B, C, H, W] to [B, HxW, C]
         x = super().forward(x.flatten(2).permute(0, 2, 1), pos=pos_embed.to(device=x.device, dtype=x.dtype))
         return x.permute(0, 2, 1).view([-1, c, h, w]).contiguous()
 
     @staticmethod
-    def build_2d_sincos_position_embedding(w, h, embed_dim=256, temperature=10000.):
-        grid_w = torch.arange(int(w), dtype=torch.float32)
-        grid_h = torch.arange(int(h), dtype=torch.float32)
-        grid_w, grid_h = torch.meshgrid(grid_w, grid_h, indexing='ij')
-        assert embed_dim % 4 == 0, \
-            'Embed dimension must be divisible by 4 for 2D sin-cos position embedding'
+    def build_2d_sincos_position_embedding(w, h, embed_dim=256, temperature=10000.0):
+        """Builds 2D sine-cosine position embedding."""
+        assert embed_dim % 4 == 0, "Embed dimension must be divisible by 4 for 2D sin-cos position embedding"
+        grid_w = torch.arange(w, dtype=torch.float32)
+        grid_h = torch.arange(h, dtype=torch.float32)
+        grid_w, grid_h = torch.meshgrid(grid_w, grid_h, indexing="ij")
         pos_dim = embed_dim // 4
         omega = torch.arange(pos_dim, dtype=torch.float32) / pos_dim
-        omega = 1. / (temperature ** omega)
+        omega = 1.0 / (temperature**omega)
 
         out_w = grid_w.flatten()[..., None] @ omega[None]
         out_h = grid_h.flatten()[..., None] @ omega[None]
@@ -140,27 +157,32 @@ class TransformerBlock(nn.Module):
 
 
 class MLPBlock(nn.Module):
+    """Implements a single block of a multi-layer perceptron."""
 
     def __init__(self, embedding_dim, mlp_dim, act=nn.GELU):
+        """Initialize the MLPBlock with specified embedding dimension, MLP dimension, and activation function."""
         super().__init__()
         self.lin1 = nn.Linear(embedding_dim, mlp_dim)
         self.lin2 = nn.Linear(mlp_dim, embedding_dim)
         self.act = act()
 
     def forward(self, x: torch.Tensor) -> torch.Tensor:
+        """Forward pass for the MLPBlock."""
         return self.lin2(self.act(self.lin1(x)))
 
 
 class MLP(nn.Module):
-    """ Very simple multi-layer perceptron (also called FFN)"""
+    """Implements a simple multi-layer perceptron (also called FFN)."""
 
     def __init__(self, input_dim, hidden_dim, output_dim, num_layers):
+        """Initialize the MLP with specified input, hidden, output dimensions and number of layers."""
         super().__init__()
         self.num_layers = num_layers
         h = [hidden_dim] * (num_layers - 1)
         self.layers = nn.ModuleList(nn.Linear(n, k) for n, k in zip([input_dim] + h, h + [output_dim]))
 
     def forward(self, x):
+        """Forward pass for the entire MLP."""
         for i, layer in enumerate(self.layers):
             x = F.relu(layer(x)) if i < self.num_layers - 1 else layer(x)
         return x
@@ -168,17 +190,23 @@ class MLP(nn.Module):
 
 class LayerNorm2d(nn.Module):
     """
-    LayerNorm2d module from https://github.com/facebookresearch/detectron2/blob/main/detectron2/layers/batch_norm.py
-    https://github.com/facebookresearch/ConvNeXt/blob/d1fa8f6fef0a165b27399986cc2bdacc92777e40/models/convnext.py#L119
+    2D Layer Normalization module inspired by Detectron2 and ConvNeXt implementations.
+
+    Original implementations in
+    https://github.com/facebookresearch/detectron2/blob/main/detectron2/layers/batch_norm.py
+    and
+    https://github.com/facebookresearch/ConvNeXt/blob/main/models/convnext.py.
     """
 
     def __init__(self, num_channels, eps=1e-6):
+        """Initialize LayerNorm2d with the given parameters."""
         super().__init__()
         self.weight = nn.Parameter(torch.ones(num_channels))
         self.bias = nn.Parameter(torch.zeros(num_channels))
         self.eps = eps
 
     def forward(self, x):
+        """Perform forward pass for 2D layer normalization."""
         u = x.mean(1, keepdim=True)
         s = (x - u).pow(2).mean(1, keepdim=True)
         x = (x - u) / torch.sqrt(s + self.eps)
@@ -187,17 +215,19 @@ class LayerNorm2d(nn.Module):
 
 class MSDeformAttn(nn.Module):
     """
-    Original Multi-Scale Deformable Attention Module.
+    Multiscale Deformable Attention Module based on Deformable-DETR and PaddleDetection implementations.
+
     https://github.com/fundamentalvision/Deformable-DETR/blob/main/models/ops/modules/ms_deform_attn.py
     """
 
     def __init__(self, d_model=256, n_levels=4, n_heads=8, n_points=4):
+        """Initialize MSDeformAttn with the given parameters."""
         super().__init__()
         if d_model % n_heads != 0:
-            raise ValueError(f'd_model must be divisible by n_heads, but got {d_model} and {n_heads}')
+            raise ValueError(f"d_model must be divisible by n_heads, but got {d_model} and {n_heads}")
         _d_per_head = d_model // n_heads
-        # you'd better set _d_per_head to a power of 2 which is more efficient in our CUDA implementation
-        assert _d_per_head * n_heads == d_model, '`d_model` must be divisible by `n_heads`'
+        # Better to set _d_per_head to a power of 2 which is more efficient in a CUDA implementation
+        assert _d_per_head * n_heads == d_model, "`d_model` must be divisible by `n_heads`"
 
         self.im2col_step = 64
 
@@ -214,25 +244,32 @@ class MSDeformAttn(nn.Module):
         self._reset_parameters()
 
     def _reset_parameters(self):
-        constant_(self.sampling_offsets.weight.data, 0.)
+        """Reset module parameters."""
+        constant_(self.sampling_offsets.weight.data, 0.0)
         thetas = torch.arange(self.n_heads, dtype=torch.float32) * (2.0 * math.pi / self.n_heads)
         grid_init = torch.stack([thetas.cos(), thetas.sin()], -1)
-        grid_init = (grid_init / grid_init.abs().max(-1, keepdim=True)[0]).view(self.n_heads, 1, 1, 2).repeat(
-            1, self.n_levels, self.n_points, 1)
+        grid_init = (
+            (grid_init / grid_init.abs().max(-1, keepdim=True)[0])
+            .view(self.n_heads, 1, 1, 2)
+            .repeat(1, self.n_levels, self.n_points, 1)
+        )
         for i in range(self.n_points):
             grid_init[:, :, i, :] *= i + 1
         with torch.no_grad():
             self.sampling_offsets.bias = nn.Parameter(grid_init.view(-1))
-        constant_(self.attention_weights.weight.data, 0.)
-        constant_(self.attention_weights.bias.data, 0.)
+        constant_(self.attention_weights.weight.data, 0.0)
+        constant_(self.attention_weights.bias.data, 0.0)
         xavier_uniform_(self.value_proj.weight.data)
-        constant_(self.value_proj.bias.data, 0.)
+        constant_(self.value_proj.bias.data, 0.0)
         xavier_uniform_(self.output_proj.weight.data)
-        constant_(self.output_proj.bias.data, 0.)
+        constant_(self.output_proj.bias.data, 0.0)
 
     def forward(self, query, refer_bbox, value, value_shapes, value_mask=None):
         """
+        Perform forward pass for multiscale deformable attention.
+
         https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/transformers/deformable_transformer.py
+
         Args:
             query (torch.Tensor): [bs, query_length, C]
             refer_bbox (torch.Tensor): [bs, query_length, n_levels, 2], range in [0, 1], top-left (0,0),
@@ -265,31 +302,34 @@ class MSDeformAttn(nn.Module):
             add = sampling_offsets / self.n_points * refer_bbox[:, :, None, :, None, 2:] * 0.5
             sampling_locations = refer_bbox[:, :, None, :, None, :2] + add
         else:
-            raise ValueError(f'Last dim of reference_points must be 2 or 4, but got {num_points}.')
+            raise ValueError(f"Last dim of reference_points must be 2 or 4, but got {num_points}.")
         output = multi_scale_deformable_attn_pytorch(value, value_shapes, sampling_locations, attention_weights)
         return self.output_proj(output)
 
 
 class DeformableTransformerDecoderLayer(nn.Module):
     """
+    Deformable Transformer Decoder Layer inspired by PaddleDetection and Deformable-DETR implementations.
+
     https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/transformers/deformable_transformer.py
     https://github.com/fundamentalvision/Deformable-DETR/blob/main/models/deformable_transformer.py
     """
 
-    def __init__(self, d_model=256, n_heads=8, d_ffn=1024, dropout=0., act=nn.ReLU(), n_levels=4, n_points=4):
+    def __init__(self, d_model=256, n_heads=8, d_ffn=1024, dropout=0.0, act=nn.ReLU(), n_levels=4, n_points=4):
+        """Initialize the DeformableTransformerDecoderLayer with the given parameters."""
         super().__init__()
 
-        # self attention
+        # Self attention
         self.self_attn = nn.MultiheadAttention(d_model, n_heads, dropout=dropout)
         self.dropout1 = nn.Dropout(dropout)
         self.norm1 = nn.LayerNorm(d_model)
 
-        # cross attention
+        # Cross attention
         self.cross_attn = MSDeformAttn(d_model, n_levels, n_heads, n_points)
         self.dropout2 = nn.Dropout(dropout)
         self.norm2 = nn.LayerNorm(d_model)
 
-        # ffn
+        # FFN
         self.linear1 = nn.Linear(d_model, d_ffn)
         self.act = act
         self.dropout3 = nn.Dropout(dropout)
@@ -299,37 +339,46 @@ class DeformableTransformerDecoderLayer(nn.Module):
 
     @staticmethod
     def with_pos_embed(tensor, pos):
+        """Add positional embeddings to the input tensor, if provided."""
         return tensor if pos is None else tensor + pos
 
     def forward_ffn(self, tgt):
+        """Perform forward pass through the Feed-Forward Network part of the layer."""
         tgt2 = self.linear2(self.dropout3(self.act(self.linear1(tgt))))
         tgt = tgt + self.dropout4(tgt2)
         return self.norm3(tgt)
 
     def forward(self, embed, refer_bbox, feats, shapes, padding_mask=None, attn_mask=None, query_pos=None):
-        # self attention
+        """Perform the forward pass through the entire decoder layer."""
+
+        # Self attention
         q = k = self.with_pos_embed(embed, query_pos)
-        tgt = self.self_attn(q.transpose(0, 1), k.transpose(0, 1), embed.transpose(0, 1),
-                             attn_mask=attn_mask)[0].transpose(0, 1)
+        tgt = self.self_attn(q.transpose(0, 1), k.transpose(0, 1), embed.transpose(0, 1), attn_mask=attn_mask)[
+            0
+        ].transpose(0, 1)
         embed = embed + self.dropout1(tgt)
         embed = self.norm1(embed)
 
-        # cross attention
-        tgt = self.cross_attn(self.with_pos_embed(embed, query_pos), refer_bbox.unsqueeze(2), feats, shapes,
-                              padding_mask)
+        # Cross attention
+        tgt = self.cross_attn(
+            self.with_pos_embed(embed, query_pos), refer_bbox.unsqueeze(2), feats, shapes, padding_mask
+        )
         embed = embed + self.dropout2(tgt)
         embed = self.norm2(embed)
 
-        # ffn
+        # FFN
         return self.forward_ffn(embed)
 
 
 class DeformableTransformerDecoder(nn.Module):
     """
+    Implementation of Deformable Transformer Decoder based on PaddleDetection.
+
     https://github.com/PaddlePaddle/PaddleDetection/blob/develop/ppdet/modeling/transformers/deformable_transformer.py
     """
 
     def __init__(self, hidden_dim, decoder_layer, num_layers, eval_idx=-1):
+        """Initialize the DeformableTransformerDecoder with the given parameters."""
         super().__init__()
         self.layers = _get_clones(decoder_layer, num_layers)
         self.num_layers = num_layers
@@ -337,16 +386,18 @@ class DeformableTransformerDecoder(nn.Module):
         self.eval_idx = eval_idx if eval_idx >= 0 else num_layers + eval_idx
 
     def forward(
-            self,
-            embed,  # decoder embeddings
-            refer_bbox,  # anchor
-            feats,  # image features
-            shapes,  # feature shapes
-            bbox_head,
-            score_head,
-            pos_mlp,
-            attn_mask=None,
-            padding_mask=None):
+        self,
+        embed,  # decoder embeddings
+        refer_bbox,  # anchor
+        feats,  # image features
+        shapes,  # feature shapes
+        bbox_head,
+        score_head,
+        pos_mlp,
+        attn_mask=None,
+        padding_mask=None,
+    ):
+        """Perform the forward pass through the entire decoder."""
         output = embed
         dec_bboxes = []
         dec_cls = []

ultralytics/nn/modules/utils.py

@@ -1,7 +1,5 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
-"""
-Module utils
-"""
+"""Module utils."""
 
 import copy
 import math
@@ -12,37 +10,44 @@ import torch.nn as nn
 import torch.nn.functional as F
 from torch.nn.init import uniform_
 
-__all__ = 'multi_scale_deformable_attn_pytorch', 'inverse_sigmoid'
+__all__ = "multi_scale_deformable_attn_pytorch", "inverse_sigmoid"
 
 
 def _get_clones(module, n):
+    """Create a list of cloned modules from the given module."""
     return nn.ModuleList([copy.deepcopy(module) for _ in range(n)])
 
 
 def bias_init_with_prob(prior_prob=0.01):
-    """initialize conv/fc bias value according to a given probability value."""
+    """Initialize conv/fc bias value according to a given probability value."""
     return float(-np.log((1 - prior_prob) / prior_prob))  # return bias_init
 
 
-def linear_init_(module):
+def linear_init(module):
+    """Initialize the weights and biases of a linear module."""
     bound = 1 / math.sqrt(module.weight.shape[0])
     uniform_(module.weight, -bound, bound)
-    if hasattr(module, 'bias') and module.bias is not None:
+    if hasattr(module, "bias") and module.bias is not None:
         uniform_(module.bias, -bound, bound)
 
 
 def inverse_sigmoid(x, eps=1e-5):
+    """Calculate the inverse sigmoid function for a tensor."""
     x = x.clamp(min=0, max=1)
     x1 = x.clamp(min=eps)
     x2 = (1 - x).clamp(min=eps)
     return torch.log(x1 / x2)
 
 
-def multi_scale_deformable_attn_pytorch(value: torch.Tensor, value_spatial_shapes: torch.Tensor,
-                                        sampling_locations: torch.Tensor,
-                                        attention_weights: torch.Tensor) -> torch.Tensor:
+def multi_scale_deformable_attn_pytorch(
+    value: torch.Tensor,
+    value_spatial_shapes: torch.Tensor,
+    sampling_locations: torch.Tensor,
+    attention_weights: torch.Tensor,
+) -> torch.Tensor:
     """
-    Multi-scale deformable attention.
+    Multiscale deformable attention.
+
     https://github.com/IDEA-Research/detrex/blob/main/detrex/layers/multi_scale_deform_attn.py
     """
 
@@ -56,23 +61,25 @@ def multi_scale_deformable_attn_pytorch(value: torch.Tensor, value_spatial_shape
         # bs, H_*W_, num_heads*embed_dims ->
         # bs, num_heads*embed_dims, H_*W_ ->
         # bs*num_heads, embed_dims, H_, W_
-        value_l_ = (value_list[level].flatten(2).transpose(1, 2).reshape(bs * num_heads, embed_dims, H_, W_))
+        value_l_ = value_list[level].flatten(2).transpose(1, 2).reshape(bs * num_heads, embed_dims, H_, W_)
         # bs, num_queries, num_heads, num_points, 2 ->
         # bs, num_heads, num_queries, num_points, 2 ->
         # bs*num_heads, num_queries, num_points, 2
         sampling_grid_l_ = sampling_grids[:, :, :, level].transpose(1, 2).flatten(0, 1)
         # bs*num_heads, embed_dims, num_queries, num_points
-        sampling_value_l_ = F.grid_sample(value_l_,
-                                          sampling_grid_l_,
-                                          mode='bilinear',
-                                          padding_mode='zeros',
-                                          align_corners=False)
+        sampling_value_l_ = F.grid_sample(
+            value_l_, sampling_grid_l_, mode="bilinear", padding_mode="zeros", align_corners=False
+        )
         sampling_value_list.append(sampling_value_l_)
     # (bs, num_queries, num_heads, num_levels, num_points) ->
     # (bs, num_heads, num_queries, num_levels, num_points) ->
     # (bs, num_heads, 1, num_queries, num_levels*num_points)
-    attention_weights = attention_weights.transpose(1, 2).reshape(bs * num_heads, 1, num_queries,
-                                                                  num_levels * num_points)
-    output = ((torch.stack(sampling_value_list, dim=-2).flatten(-2) * attention_weights).sum(-1).view(
-        bs, num_heads * embed_dims, num_queries))
+    attention_weights = attention_weights.transpose(1, 2).reshape(
+        bs * num_heads, 1, num_queries, num_levels * num_points
+    )
+    output = (
+        (torch.stack(sampling_value_list, dim=-2).flatten(-2) * attention_weights)
+        .sum(-1)
+        .view(bs, num_heads * embed_dims, num_queries)
+    )
     return output.transpose(1, 2).contiguous()