ultralytics 8.2.103__py3-none-any.whl → 8.3.0__py3-none-any.whl

ultralytics/data/augment.py

@@ -1628,92 +1628,105 @@ class LetterBox:
         return labels
 
 
-class CopyPaste:
+class CopyPaste(BaseMixTransform):
     """
-    Implements Copy-Paste augmentation as described in https://arxiv.org/abs/2012.07177.
+    CopyPaste class for applying Copy-Paste augmentation to image datasets.
 
-    This class applies Copy-Paste augmentation on images and their corresponding instances.
+    This class implements the Copy-Paste augmentation technique as described in the paper "Simple Copy-Paste is a Strong
+    Data Augmentation Method for Instance Segmentation" (https://arxiv.org/abs/2012.07177). It combines objects from
+    different images to create new training samples.
 
     Attributes:
-        p (float): Probability of applying the Copy-Paste augmentation. Must be between 0 and 1.
+        dataset (Any): The dataset to which Copy-Paste augmentation will be applied.
+        pre_transform (Callable | None): Optional transform to apply before Copy-Paste.
+        p (float): Probability of applying Copy-Paste augmentation.
 
     Methods:
-        __call__: Applies Copy-Paste augmentation to given image and instances.
+        get_indexes: Returns a random index from the dataset.
+        _mix_transform: Applies Copy-Paste augmentation to the input labels.
+        __call__: Applies the Copy-Paste transformation to images and annotations.
 
     Examples:
-        >>> copypaste = CopyPaste(p=0.5)
-        >>> augmented_labels = copypaste(labels)
-        >>> augmented_image = augmented_labels["img"]
+        >>> from ultralytics.data.augment import CopyPaste
+        >>> dataset = YourDataset(...)  # Your image dataset
+        >>> copypaste = CopyPaste(dataset, p=0.5)
+        >>> augmented_labels = copypaste(original_labels)
     """
 
-    def __init__(self, p=0.5) -> None:
-        """
-        Initializes the CopyPaste augmentation object.
+    def __init__(self, dataset=None, pre_transform=None, p=0.5, mode="flip") -> None:
+        """Initializes CopyPaste object with dataset, pre_transform, and probability of applying MixUp."""
+        super().__init__(dataset=dataset, pre_transform=pre_transform, p=p)
+        assert mode in {"flip", "mixup"}, f"Expected `mode` to be `flip` or `mixup`, but got {mode}."
+        self.mode = mode
 
-        This class implements the Copy-Paste augmentation as described in the paper "Simple Copy-Paste is a Strong Data
-        Augmentation Method for Instance Segmentation" (https://arxiv.org/abs/2012.07177). It applies the Copy-Paste
-        augmentation on images and their corresponding instances with a given probability.
+    def get_indexes(self):
+        """Returns a list of random indexes from the dataset for CopyPaste augmentation."""
+        return random.randint(0, len(self.dataset) - 1)
 
-        Args:
-            p (float): The probability of applying the Copy-Paste augmentation. Must be between 0 and 1.
+    def _mix_transform(self, labels):
+        """Applies Copy-Paste augmentation to combine objects from another image into the current image."""
+        labels2 = labels["mix_labels"][0]
+        return self._transform(labels, labels2)
 
-        Attributes:
-            p (float): Stores the probability of applying the augmentation.
+    def __call__(self, labels):
+        """Applies Copy-Paste augmentation to an image and its labels."""
+        if len(labels["instances"].segments) == 0 or self.p == 0:
+            return labels
+        if self.mode == "flip":
+            return self._transform(labels)
 
-        Examples:
-            >>> augment = CopyPaste(p=0.7)
-            >>> augmented_data = augment(original_data)
-        """
-        self.p = p
+        # Get index of one or three other images
+        indexes = self.get_indexes()
+        if isinstance(indexes, int):
+            indexes = [indexes]
 
-    def __call__(self, labels):
-        """
-        Applies Copy-Paste augmentation to an image and its instances.
+        # Get images information will be used for Mosaic or MixUp
+        mix_labels = [self.dataset.get_image_and_label(i) for i in indexes]
 
-        Args:
-            labels (Dict): A dictionary containing:
-                - 'img' (np.ndarray): The image to augment.
-                - 'cls' (np.ndarray): Class labels for the instances.
-                - 'instances' (ultralytics.engine.results.Instances): Object containing bounding boxes, segments, etc.
+        if self.pre_transform is not None:
+            for i, data in enumerate(mix_labels):
+                mix_labels[i] = self.pre_transform(data)
+        labels["mix_labels"] = mix_labels
 
-        Returns:
-            (Dict): Dictionary with augmented image and updated instances under 'img', 'cls', and 'instances' keys.
+        # Update cls and texts
+        labels = self._update_label_text(labels)
+        # Mosaic or MixUp
+        labels = self._mix_transform(labels)
+        labels.pop("mix_labels", None)
+        return labels
 
-        Examples:
-            >>> labels = {"img": np.random.rand(640, 640, 3), "cls": np.array([0, 1, 2]), "instances": Instances(...)}
-            >>> augmenter = CopyPaste(p=0.5)
-            >>> augmented_labels = augmenter(labels)
-        """
-        im = labels["img"]
-        cls = labels["cls"]
+    def _transform(self, labels1, labels2={}):
+        """Applies Copy-Paste augmentation to combine objects from another image into the current image."""
+        im = labels1["img"]
+        cls = labels1["cls"]
         h, w = im.shape[:2]
-        instances = labels.pop("instances")
+        instances = labels1.pop("instances")
         instances.convert_bbox(format="xyxy")
         instances.denormalize(w, h)
-        if self.p and len(instances.segments):
-            _, w, _ = im.shape  # height, width, channels
-            im_new = np.zeros(im.shape, np.uint8)
-
-            # Calculate ioa first then select indexes randomly
-            ins_flip = deepcopy(instances)
-            ins_flip.fliplr(w)
-
-            ioa = bbox_ioa(ins_flip.bboxes, instances.bboxes)  # intersection over area, (N, M)
-            indexes = np.nonzero((ioa < 0.30).all(1))[0]  # (N, )
-            n = len(indexes)
-            for j in random.sample(list(indexes), k=round(self.p * n)):
-                cls = np.concatenate((cls, cls[[j]]), axis=0)
-                instances = Instances.concatenate((instances, ins_flip[[j]]), axis=0)
-                cv2.drawContours(im_new, instances.segments[[j]].astype(np.int32), -1, (1, 1, 1), cv2.FILLED)
-
-            result = cv2.flip(im, 1)  # augment segments (flip left-right)
-            i = cv2.flip(im_new, 1).astype(bool)
-            im[i] = result[i]
-
-        labels["img"] = im
-        labels["cls"] = cls
-        labels["instances"] = instances
-        return labels
+
+        im_new = np.zeros(im.shape, np.uint8)
+        instances2 = labels2.pop("instances", None)
+        if instances2 is None:
+            instances2 = deepcopy(instances)
+            instances2.fliplr(w)
+        ioa = bbox_ioa(instances2.bboxes, instances.bboxes)  # intersection over area, (N, M)
+        indexes = np.nonzero((ioa < 0.30).all(1))[0]  # (N, )
+        n = len(indexes)
+        sorted_idx = np.argsort(ioa.max(1)[indexes])
+        indexes = indexes[sorted_idx]
+        for j in indexes[: round(self.p * n)]:
+            cls = np.concatenate((cls, labels2.get("cls", cls)[[j]]), axis=0)
+            instances = Instances.concatenate((instances, instances2[[j]]), axis=0)
+            cv2.drawContours(im_new, instances2.segments[[j]].astype(np.int32), -1, (1, 1, 1), cv2.FILLED)
+
+        result = labels2.get("img", cv2.flip(im, 1))  # augment segments
+        i = im_new.astype(bool)
+        im[i] = result[i]
+
+        labels1["img"] = im
+        labels1["cls"] = cls
+        labels1["instances"] = instances
+        return labels1
 
 
 class Albumentations:
@@ -2259,9 +2272,9 @@ class RandomLoadText:
 
 def v8_transforms(dataset, imgsz, hyp, stretch=False):
     """
-    Applies a series of image transformations for YOLOv8 training.
+    Applies a series of image transformations for training.
 
-    This function creates a composition of image augmentation techniques to prepare images for YOLOv8 training.
+    This function creates a composition of image augmentation techniques to prepare images for YOLO training.
     It includes operations such as mosaic, copy-paste, random perspective, mixup, and various color adjustments.
 
     Args:
@@ -2280,20 +2293,28 @@ def v8_transforms(dataset, imgsz, hyp, stretch=False):
         >>> transforms = v8_transforms(dataset, imgsz=640, hyp=hyp)
         >>> augmented_data = transforms(dataset[0])
     """
-    pre_transform = Compose(
-        [
-            Mosaic(dataset, imgsz=imgsz, p=hyp.mosaic),
-            CopyPaste(p=hyp.copy_paste),
-            RandomPerspective(
-                degrees=hyp.degrees,
-                translate=hyp.translate,
-                scale=hyp.scale,
-                shear=hyp.shear,
-                perspective=hyp.perspective,
-                pre_transform=None if stretch else LetterBox(new_shape=(imgsz, imgsz)),
-            ),
-        ]
+    mosaic = Mosaic(dataset, imgsz=imgsz, p=hyp.mosaic)
+    affine = RandomPerspective(
+        degrees=hyp.degrees,
+        translate=hyp.translate,
+        scale=hyp.scale,
+        shear=hyp.shear,
+        perspective=hyp.perspective,
+        pre_transform=None if stretch else LetterBox(new_shape=(imgsz, imgsz)),
     )
+
+    pre_transform = Compose([mosaic, affine])
+    if hyp.copy_paste_mode == "flip":
+        pre_transform.insert(1, CopyPaste(p=hyp.copy_paste, mode=hyp.copy_paste_mode))
+    else:
+        pre_transform.append(
+            CopyPaste(
+                dataset,
+                pre_transform=Compose([Mosaic(dataset, imgsz=imgsz, p=hyp.mosaic), affine]),
+                p=hyp.copy_paste,
+                mode=hyp.copy_paste_mode,
+            )
+        )
     flip_idx = dataset.data.get("flip_idx", [])  # for keypoints augmentation
     if dataset.use_keypoints:
         kpt_shape = dataset.data.get("kpt_shape", None)

ultralytics/engine/trainer.py

@@ -538,6 +538,8 @@ class BaseTrainer:
             self.best.write_bytes(serialized_ckpt)  # save best.pt
         if (self.save_period > 0) and (self.epoch % self.save_period == 0):
             (self.wdir / f"epoch{self.epoch}.pt").write_bytes(serialized_ckpt)  # save epoch, i.e. 'epoch3.pt'
+        # if self.args.close_mosaic and self.epoch == (self.epochs - self.args.close_mosaic - 1):
+        #    (self.wdir / "last_mosaic.pt").write_bytes(serialized_ckpt)  # save mosaic checkpoint
 
     def get_dataset(self):
         """
@@ -698,7 +700,12 @@ class BaseTrainer:
                 resume = True
                 self.args = get_cfg(ckpt_args)
                 self.args.model = self.args.resume = str(last)  # reinstate model
-                for k in "imgsz", "batch", "device":  # allow arg updates to reduce memory or update device on resume
+                for k in (
+                    "imgsz",
+                    "batch",
+                    "device",
+                    "close_mosaic",
+                ):  # allow arg updates to reduce memory or update device on resume
                     if k in overrides:
                         setattr(self.args, k, overrides[k])
 

ultralytics/nn/modules/__init__.py

@@ -20,6 +20,7 @@ Example:
 from .block import (
     C1,
     C2,
+    C2PSA,
     C3,
     C3TR,
     CIB,
@@ -38,7 +39,9 @@ from .block import (
     C2f,
     C2fAttn,
     C2fCIB,
+    C2fPSA,
     C3Ghost,
+    C3k2,
     C3x,
     CBFuse,
     CBLinear,
@@ -110,6 +113,10 @@ __all__ = (
     "C2",
     "C3",
     "C2f",
+    "C3k2",
+    "SCDown",
+    "C2fPSA",
+    "C2PSA",
     "C2fAttn",
     "C3x",
     "C3TR",
@@ -149,5 +156,4 @@ __all__ = (
     "C2fCIB",
     "Attention",
     "PSA",
-    "SCDown",
 )

ultralytics/nn/modules/block.py

@@ -40,6 +40,9 @@ __all__ = (
     "SPPELAN",
     "CBFuse",
     "CBLinear",
+    "C3k2",
+    "C2fPSA",
+    "C2PSA",
     "RepVGGDW",
     "CIB",
     "C2fCIB",
@@ -696,6 +699,49 @@ class CBFuse(nn.Module):
         return torch.sum(torch.stack(res + xs[-1:]), dim=0)
 
 
+class C3f(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__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c1, c_, 1, 1)
+        self.cv3 = Conv((2 + n) * c_, c2, 1)  # optional act=FReLU(c2)
+        self.m = nn.ModuleList(Bottleneck(c_, 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 = [self.cv2(x), self.cv1(x)]
+        y.extend(m(y[-1]) for m in self.m)
+        return self.cv3(torch.cat(y, 1))
+
+
+class C3k2(C2f):
+    """Faster Implementation of CSP Bottleneck with 2 convolutions."""
+
+    def __init__(self, c1, c2, n=1, c3k=False, e=0.5, g=1, shortcut=True):
+        """Initializes the C3k2 module, a faster CSP Bottleneck with 2 convolutions and optional C3k blocks."""
+        super().__init__(c1, c2, n, shortcut, g, e)
+        self.m = nn.ModuleList(
+            C3k(self.c, self.c, 2, shortcut, g) if c3k else Bottleneck(self.c, self.c, shortcut, g) for _ in range(n)
+        )
+
+
+class C3k(C3):
+    """C3k is a CSP bottleneck module with customizable kernel sizes for feature extraction in neural networks."""
+
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5, k=3):
+        """Initializes the C3k module with specified channels, number of layers, and configurations."""
+        super().__init__(c1, c2, n, shortcut, g, e)
+        c_ = int(c2 * e)  # hidden channels
+        # self.m = nn.Sequential(*(RepBottleneck(c_, c_, shortcut, g, k=(k, k), e=1.0) for _ in range(n)))
+        self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, k=(k, k), e=1.0) for _ in range(n)))
+
+
 class RepVGGDW(torch.nn.Module):
     """RepVGGDW is a class that represents a depth wise separable convolutional block in RepVGG architecture."""
 
@@ -873,25 +919,69 @@ class Attention(nn.Module):
         return x
 
 
+class PSABlock(nn.Module):
+    """
+    PSABlock class implementing a Position-Sensitive Attention block for neural networks.
+
+    This class encapsulates the functionality for applying multi-head attention and feed-forward neural network layers
+    with optional shortcut connections.
+
+    Attributes:
+        attn (Attention): Multi-head attention module.
+        ffn (nn.Sequential): Feed-forward neural network module.
+        add (bool): Flag indicating whether to add shortcut connections.
+
+    Methods:
+        forward: Performs a forward pass through the PSABlock, applying attention and feed-forward layers.
+
+    Examples:
+        Create a PSABlock and perform a forward pass
+        >>> psablock = PSABlock(c=128, attn_ratio=0.5, num_heads=4, shortcut=True)
+        >>> input_tensor = torch.randn(1, 128, 32, 32)
+        >>> output_tensor = psablock(input_tensor)
+    """
+
+    def __init__(self, c, attn_ratio=0.5, num_heads=4, shortcut=True) -> None:
+        """Initializes the PSABlock with attention and feed-forward layers for enhanced feature extraction."""
+        super().__init__()
+
+        self.attn = Attention(c, attn_ratio=attn_ratio, num_heads=num_heads)
+        self.ffn = nn.Sequential(Conv(c, c * 2, 1), Conv(c * 2, c, 1, act=False))
+        self.add = shortcut
+
+    def forward(self, x):
+        """Executes a forward pass through PSABlock, applying attention and feed-forward layers to the input tensor."""
+        x = x + self.attn(x) if self.add else self.attn(x)
+        x = x + self.ffn(x) if self.add else self.ffn(x)
+        return x
+
+
 class PSA(nn.Module):
     """
-    Position-wise Spatial Attention module.
+    PSA class for implementing Position-Sensitive Attention in neural networks.
 
-    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.
+    This class encapsulates the functionality for applying position-sensitive attention and feed-forward networks to
+    input tensors, enhancing feature extraction and processing capabilities.
 
     Attributes:
-        c (int): Number of intermediate channels.
+        c (int): Number of hidden channels after applying the initial convolution.
         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.
+        attn (Attention): Attention module for position-sensitive attention.
+        ffn (nn.Sequential): Feed-forward network for further processing.
+
+    Methods:
+        forward: Applies position-sensitive attention and feed-forward network to the input tensor.
+
+    Examples:
+        Create a PSA module and apply it to an input tensor
+        >>> psa = PSA(c1=128, c2=128, e=0.5)
+        >>> input_tensor = torch.randn(1, 128, 64, 64)
+        >>> output_tensor = psa.forward(input_tensor)
     """
 
     def __init__(self, c1, c2, e=0.5):
-        """Initializes convolution layers, attention module, and feed-forward network with channel reduction."""
+        """Initializes the PSA module with input/output channels and attention mechanism for feature extraction."""
         super().__init__()
         assert c1 == c2
         self.c = int(c1 * e)
@@ -902,46 +992,117 @@ class PSA(nn.Module):
         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.
-        """
+        """Executes forward pass in PSA module, applying attention and feed-forward layers to the input 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))
 
 
+class C2PSA(nn.Module):
+    """
+    C2PSA module with attention mechanism for enhanced feature extraction and processing.
+
+    This module implements a convolutional block with attention mechanisms to enhance feature extraction and processing
+    capabilities. It includes a series of PSABlock modules for self-attention and feed-forward operations.
+
+    Attributes:
+        c (int): Number of hidden 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.
+        m (nn.Sequential): Sequential container of PSABlock modules for attention and feed-forward operations.
+
+    Methods:
+        forward: Performs a forward pass through the C2PSA module, applying attention and feed-forward operations.
+
+    Notes:
+        This module essentially is the same as PSA module, but refactored to allow stacking more PSABlock modules.
+
+    Examples:
+        >>> c2psa = C2PSA(c1=256, c2=256, n=3, e=0.5)
+        >>> input_tensor = torch.randn(1, 256, 64, 64)
+        >>> output_tensor = c2psa(input_tensor)
+    """
+
+    def __init__(self, c1, c2, n=1, e=0.5):
+        """Initializes the C2PSA module with specified input/output channels, number of layers, and expansion ratio."""
+        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.m = nn.Sequential(*(PSABlock(self.c, attn_ratio=0.5, num_heads=self.c // 64) for _ in range(n)))
+
+    def forward(self, x):
+        """Processes the input tensor 'x' through a series of PSA blocks and returns the transformed tensor."""
+        a, b = self.cv1(x).split((self.c, self.c), dim=1)
+        b = self.m(b)
+        return self.cv2(torch.cat((a, b), 1))
+
+
+class C2fPSA(C2f):
+    """
+    C2fPSA module with enhanced feature extraction using PSA blocks.
+
+    This class extends the C2f module by incorporating PSA blocks for improved attention mechanisms and feature extraction.
+
+    Attributes:
+        c (int): Number of hidden 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.
+        m (nn.ModuleList): List of PSA blocks for feature extraction.
+
+    Methods:
+        forward: Performs a forward pass through the C2fPSA module.
+        forward_split: Performs a forward pass using split() instead of chunk().
+
+    Examples:
+        >>> import torch
+        >>> from ultralytics.models.common import C2fPSA
+        >>> model = C2fPSA(c1=64, c2=64, n=3, e=0.5)
+        >>> x = torch.randn(1, 64, 128, 128)
+        >>> output = model(x)
+        >>> print(output.shape)
+    """
+
+    def __init__(self, c1, c2, n=1, e=0.5):
+        """Initializes the C2fPSA module, a variant of C2f with PSA blocks for enhanced feature extraction."""
+        assert c1 == c2
+        super().__init__(c1, c2, n=n, e=e)
+        self.m = nn.ModuleList(PSABlock(self.c, attn_ratio=0.5, num_heads=self.c // 64) for _ in range(n))
+
+
 class SCDown(nn.Module):
-    """Spatial Channel Downsample (SCDown) module for reducing spatial and channel dimensions."""
+    """
+    SCDown module for downsampling with separable convolutions.
 
-    def __init__(self, c1, c2, k, s):
-        """
-        Spatial Channel Downsample (SCDown) module.
+    This module performs downsampling using a combination of pointwise and depthwise convolutions, which helps in
+    efficiently reducing the spatial dimensions of the input tensor while maintaining the channel information.
 
-        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.
-        """
+    Attributes:
+        cv1 (Conv): Pointwise convolution layer that reduces the number of channels.
+        cv2 (Conv): Depthwise convolution layer that performs spatial downsampling.
+
+    Methods:
+        forward: Applies the SCDown module to the input tensor.
+
+    Examples:
+        >>> import torch
+        >>> from ultralytics import SCDown
+        >>> model = SCDown(c1=64, c2=128, k=3, s=2)
+        >>> x = torch.randn(1, 64, 128, 128)
+        >>> y = model(x)
+        >>> print(y.shape)
+        torch.Size([1, 128, 64, 64])
+    """
+
+    def __init__(self, c1, c2, k, s):
+        """Initializes the SCDown module with specified input/output channels, kernel size, and stride."""
         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.
-        """
+        """Applies convolution and downsampling to the input tensor in the SCDown module."""
         return self.cv2(self.cv1(x))

ultralytics/nn/modules/conv.py

@@ -209,7 +209,8 @@ class RepConv(nn.Module):
         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):
+    @staticmethod
+    def _pad_1x1_to_3x3_tensor(kernel1x1):
         """Pads a 1x1 tensor to a 3x3 tensor."""
         if kernel1x1 is None:
             return 0

ultralytics/nn/modules/head.py

@@ -11,7 +11,7 @@ from torch.nn.init import constant_, xavier_uniform_
 from ultralytics.utils.tal import TORCH_1_10, dist2bbox, dist2rbox, make_anchors
 
 from .block import DFL, BNContrastiveHead, ContrastiveHead, Proto
-from .conv import Conv
+from .conv import Conv, DWConv
 from .transformer import MLP, DeformableTransformerDecoder, DeformableTransformerDecoderLayer
 from .utils import bias_init_with_prob, linear_init
 
@@ -41,7 +41,14 @@ class Detect(nn.Module):
         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
         )
-        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.cv3 = nn.ModuleList(
+            nn.Sequential(
+                nn.Sequential(DWConv(x, x, 3), Conv(x, c3, 1)),
+                nn.Sequential(DWConv(c3, c3, 3), Conv(c3, c3, 1)),
+                nn.Conv2d(c3, self.nc, 1),
+            )
+            for x in ch
+        )
         self.dfl = DFL(self.reg_max) if self.reg_max > 1 else nn.Identity()
 
         if self.end2end: