ultralytics 8.3.189__py3-none-any.whl → 8.3.191__py3-none-any.whl

ultralytics/nn/modules/block.py

@@ -1,7 +1,7 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 """Block modules."""
 
-from typing import List, Optional, Tuple
+from __future__ import annotations
 
 import torch
 import torch.nn as nn
@@ -192,7 +192,7 @@ class HGBlock(nn.Module):
 class SPP(nn.Module):
     """Spatial Pyramid Pooling (SPP) layer https://arxiv.org/abs/1406.4729."""
 
-    def __init__(self, c1: int, c2: int, k: Tuple[int, ...] = (5, 9, 13)):
+    def __init__(self, c1: int, c2: int, k: tuple[int, ...] = (5, 9, 13)):
         """
         Initialize the SPP layer with input/output channels and pooling kernel sizes.
 
@@ -471,7 +471,7 @@ class Bottleneck(nn.Module):
     """Standard bottleneck."""
 
     def __init__(
-        self, c1: int, c2: int, shortcut: bool = True, g: int = 1, k: Tuple[int, int] = (3, 3), e: float = 0.5
+        self, c1: int, c2: int, shortcut: bool = True, g: int = 1, k: tuple[int, int] = (3, 3), e: float = 0.5
     ):
         """
         Initialize a standard bottleneck module.
@@ -711,7 +711,7 @@ class ImagePoolingAttn(nn.Module):
     """ImagePoolingAttn: Enhance the text embeddings with image-aware information."""
 
     def __init__(
-        self, ec: int = 256, ch: Tuple[int, ...] = (), ct: int = 512, nh: int = 8, k: int = 3, scale: bool = False
+        self, ec: int = 256, ch: tuple[int, ...] = (), ct: int = 512, nh: int = 8, k: int = 3, scale: bool = False
     ):
         """
         Initialize ImagePoolingAttn module.
@@ -740,7 +740,7 @@ class ImagePoolingAttn(nn.Module):
         self.hc = ec // nh
         self.k = k
 
-    def forward(self, x: List[torch.Tensor], text: torch.Tensor) -> torch.Tensor:
+    def forward(self, x: list[torch.Tensor], text: torch.Tensor) -> torch.Tensor:
         """
         Forward pass of ImagePoolingAttn.
 
@@ -856,7 +856,7 @@ class RepBottleneck(Bottleneck):
     """Rep bottleneck."""
 
     def __init__(
-        self, c1: int, c2: int, shortcut: bool = True, g: int = 1, k: Tuple[int, int] = (3, 3), e: float = 0.5
+        self, c1: int, c2: int, shortcut: bool = True, g: int = 1, k: tuple[int, int] = (3, 3), e: float = 0.5
     ):
         """
         Initialize RepBottleneck.
@@ -1026,7 +1026,7 @@ class SPPELAN(nn.Module):
 class CBLinear(nn.Module):
     """CBLinear."""
 
-    def __init__(self, c1: int, c2s: List[int], k: int = 1, s: int = 1, p: Optional[int] = None, g: int = 1):
+    def __init__(self, c1: int, c2s: list[int], k: int = 1, s: int = 1, p: int | None = None, g: int = 1):
         """
         Initialize CBLinear module.
 
@@ -1042,7 +1042,7 @@ class CBLinear(nn.Module):
         self.c2s = c2s
         self.conv = nn.Conv2d(c1, sum(c2s), k, s, autopad(k, p), groups=g, bias=True)
 
-    def forward(self, x: torch.Tensor) -> List[torch.Tensor]:
+    def forward(self, x: torch.Tensor) -> list[torch.Tensor]:
         """Forward pass through CBLinear layer."""
         return self.conv(x).split(self.c2s, dim=1)
 
@@ -1050,7 +1050,7 @@ class CBLinear(nn.Module):
 class CBFuse(nn.Module):
     """CBFuse."""
 
-    def __init__(self, idx: List[int]):
+    def __init__(self, idx: list[int]):
         """
         Initialize CBFuse module.
 
@@ -1060,7 +1060,7 @@ class CBFuse(nn.Module):
         super().__init__()
         self.idx = idx
 
-    def forward(self, xs: List[torch.Tensor]) -> torch.Tensor:
+    def forward(self, xs: list[torch.Tensor]) -> torch.Tensor:
         """
         Forward pass through CBFuse layer.
 
@@ -1974,7 +1974,7 @@ class Residual(nn.Module):
 class SAVPE(nn.Module):
     """Spatial-Aware Visual Prompt Embedding module for feature enhancement."""
 
-    def __init__(self, ch: List[int], c3: int, embed: int):
+    def __init__(self, ch: list[int], c3: int, embed: int):
         """
         Initialize SAVPE module with channels, intermediate channels, and embedding dimension.
 
@@ -2002,7 +2002,7 @@ class SAVPE(nn.Module):
         self.cv5 = nn.Conv2d(1, self.c, 3, padding=1)
         self.cv6 = nn.Sequential(Conv(2 * self.c, self.c, 3), nn.Conv2d(self.c, self.c, 3, padding=1))
 
-    def forward(self, x: List[torch.Tensor], vp: torch.Tensor) -> torch.Tensor:
+    def forward(self, x: list[torch.Tensor], vp: torch.Tensor) -> torch.Tensor:
         """Process input features and visual prompts to generate enhanced embeddings."""
         y = [self.cv2[i](xi) for i, xi in enumerate(x)]
         y = self.cv4(torch.cat(y, dim=1))

ultralytics/nn/modules/conv.py

@@ -1,8 +1,9 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 """Convolution modules."""
 
+from __future__ import annotations
+
 import math
-from typing import List
 
 import numpy as np
 import torch
@@ -669,7 +670,7 @@ class Concat(nn.Module):
         super().__init__()
         self.d = dimension
 
-    def forward(self, x: List[torch.Tensor]):
+    def forward(self, x: list[torch.Tensor]):
         """
         Concatenate input tensors along specified dimension.
 
@@ -700,7 +701,7 @@ class Index(nn.Module):
         super().__init__()
         self.index = index
 
-    def forward(self, x: List[torch.Tensor]):
+    def forward(self, x: list[torch.Tensor]):
         """
         Select and return a particular index from input.
 

ultralytics/nn/modules/head.py

@@ -1,15 +1,17 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 """Model head modules."""
 
+from __future__ import annotations
+
 import copy
 import math
-from typing import List, Optional, Tuple, Union
 
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 from torch.nn.init import constant_, xavier_uniform_
 
+from ultralytics.utils import NOT_MACOS14
 from ultralytics.utils.tal import TORCH_1_10, dist2bbox, dist2rbox, make_anchors
 from ultralytics.utils.torch_utils import fuse_conv_and_bn, smart_inference_mode
 
@@ -75,7 +77,7 @@ class Detect(nn.Module):
     legacy = False  # backward compatibility for v3/v5/v8/v9 models
     xyxy = False  # xyxy or xywh output
 
-    def __init__(self, nc: int = 80, ch: Tuple = ()):
+    def __init__(self, nc: int = 80, ch: tuple = ()):
         """
         Initialize the YOLO detection layer with specified number of classes and channels.
 
@@ -111,7 +113,7 @@ class Detect(nn.Module):
             self.one2one_cv2 = copy.deepcopy(self.cv2)
             self.one2one_cv3 = copy.deepcopy(self.cv3)
 
-    def forward(self, x: List[torch.Tensor]) -> Union[List[torch.Tensor], Tuple]:
+    def forward(self, x: list[torch.Tensor]) -> list[torch.Tensor] | tuple:
         """Concatenate and return predicted bounding boxes and class probabilities."""
         if self.end2end:
             return self.forward_end2end(x)
@@ -123,7 +125,7 @@ class Detect(nn.Module):
         y = self._inference(x)
         return y if self.export else (y, x)
 
-    def forward_end2end(self, x: List[torch.Tensor]) -> Union[dict, Tuple]:
+    def forward_end2end(self, x: list[torch.Tensor]) -> dict | tuple:
         """
         Perform forward pass of the v10Detect module.
 
@@ -147,7 +149,7 @@ class Detect(nn.Module):
         y = self.postprocess(y.permute(0, 2, 1), self.max_det, self.nc)
         return y if self.export else (y, {"one2many": x, "one2one": one2one})
 
-    def _inference(self, x: List[torch.Tensor]) -> torch.Tensor:
+    def _inference(self, x: list[torch.Tensor]) -> torch.Tensor:
         """
         Decode predicted bounding boxes and class probabilities based on multiple-level feature maps.
 
@@ -199,7 +201,12 @@ class Detect(nn.Module):
 
     def decode_bboxes(self, bboxes: torch.Tensor, anchors: torch.Tensor, xywh: bool = True) -> torch.Tensor:
         """Decode bounding boxes from predictions."""
-        return dist2bbox(bboxes, anchors, xywh=xywh and not (self.end2end or self.xyxy), dim=1)
+        return dist2bbox(
+            bboxes,
+            anchors,
+            xywh=xywh and not self.end2end and not self.xyxy,
+            dim=1,
+        )
 
     @staticmethod
     def postprocess(preds: torch.Tensor, max_det: int, nc: int = 80) -> torch.Tensor:
@@ -248,7 +255,7 @@ class Segment(Detect):
         >>> outputs = segment(x)
     """
 
-    def __init__(self, nc: int = 80, nm: int = 32, npr: int = 256, ch: Tuple = ()):
+    def __init__(self, nc: int = 80, nm: int = 32, npr: int = 256, ch: tuple = ()):
         """
         Initialize the YOLO model attributes such as the number of masks, prototypes, and the convolution layers.
 
@@ -266,7 +273,7 @@ class Segment(Detect):
         c4 = max(ch[0] // 4, self.nm)
         self.cv4 = nn.ModuleList(nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3), nn.Conv2d(c4, self.nm, 1)) for x in ch)
 
-    def forward(self, x: List[torch.Tensor]) -> Union[Tuple, List[torch.Tensor]]:
+    def forward(self, x: list[torch.Tensor]) -> tuple | list[torch.Tensor]:
         """Return model outputs and mask coefficients if training, otherwise return outputs and mask coefficients."""
         p = self.proto(x[0])  # mask protos
         bs = p.shape[0]  # batch size
@@ -300,7 +307,7 @@ class OBB(Detect):
         >>> outputs = obb(x)
     """
 
-    def __init__(self, nc: int = 80, ne: int = 1, ch: Tuple = ()):
+    def __init__(self, nc: int = 80, ne: int = 1, ch: tuple = ()):
         """
         Initialize OBB with number of classes `nc` and layer channels `ch`.
 
@@ -315,7 +322,7 @@ class OBB(Detect):
         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: List[torch.Tensor]) -> Union[torch.Tensor, Tuple]:
+    def forward(self, x: list[torch.Tensor]) -> torch.Tensor | tuple:
         """Concatenate and return 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
@@ -356,7 +363,7 @@ class Pose(Detect):
         >>> outputs = pose(x)
     """
 
-    def __init__(self, nc: int = 80, kpt_shape: Tuple = (17, 3), ch: Tuple = ()):
+    def __init__(self, nc: int = 80, kpt_shape: tuple = (17, 3), ch: tuple = ()):
         """
         Initialize YOLO network with default parameters and Convolutional Layers.
 
@@ -372,7 +379,7 @@ class Pose(Detect):
         c4 = max(ch[0] // 4, self.nk)
         self.cv4 = nn.ModuleList(nn.Sequential(Conv(x, c4, 3), Conv(c4, c4, 3), nn.Conv2d(c4, self.nk, 1)) for x in ch)
 
-    def forward(self, x: List[torch.Tensor]) -> Union[torch.Tensor, Tuple]:
+    def forward(self, x: list[torch.Tensor]) -> torch.Tensor | tuple:
         """Perform forward pass through YOLO model and return predictions."""
         bs = x[0].shape[0]  # batch size
         kpt = torch.cat([self.cv4[i](x[i]).view(bs, self.nk, -1) for i in range(self.nl)], -1)  # (bs, 17*3, h*w)
@@ -408,7 +415,10 @@ class Pose(Detect):
         else:
             y = kpts.clone()
             if ndim == 3:
-                y[:, 2::ndim] = y[:, 2::ndim].sigmoid()  # sigmoid (WARNING: inplace .sigmoid_() Apple MPS bug)
+                if NOT_MACOS14:
+                    y[:, 2::ndim].sigmoid_()
+                else:  # Apple macOS14 MPS bug https://github.com/ultralytics/ultralytics/pull/21878
+                    y[:, 2::ndim] = y[:, 2::ndim].sigmoid()
             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
@@ -439,7 +449,7 @@ class Classify(nn.Module):
 
     export = False  # export mode
 
-    def __init__(self, c1: int, c2: int, k: int = 1, s: int = 1, p: Optional[int] = None, g: int = 1):
+    def __init__(self, c1: int, c2: int, k: int = 1, s: int = 1, p: int | None = None, g: int = 1):
         """
         Initialize YOLO classification head to transform input tensor from (b,c1,20,20) to (b,c2) shape.
 
@@ -458,7 +468,7 @@ class Classify(nn.Module):
         self.drop = nn.Dropout(p=0.0, inplace=True)
         self.linear = nn.Linear(c_, c2)  # to x(b,c2)
 
-    def forward(self, x: Union[List[torch.Tensor], torch.Tensor]) -> Union[torch.Tensor, Tuple]:
+    def forward(self, x: list[torch.Tensor] | torch.Tensor) -> torch.Tensor | tuple:
         """Perform forward pass of the YOLO model on input image data."""
         if isinstance(x, list):
             x = torch.cat(x, 1)
@@ -492,7 +502,7 @@ class WorldDetect(Detect):
         >>> outputs = world_detect(x, text)
     """
 
-    def __init__(self, nc: int = 80, embed: int = 512, with_bn: bool = False, ch: Tuple = ()):
+    def __init__(self, nc: int = 80, embed: int = 512, with_bn: bool = False, ch: tuple = ()):
         """
         Initialize YOLO detection layer with nc classes and layer channels ch.
 
@@ -507,7 +517,7 @@ class WorldDetect(Detect):
         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: List[torch.Tensor], text: torch.Tensor) -> Union[List[torch.Tensor], Tuple]:
+    def forward(self, x: list[torch.Tensor], text: torch.Tensor) -> list[torch.Tensor] | tuple:
         """Concatenate and return 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)
@@ -576,7 +586,7 @@ class LRPCHead(nn.Module):
         linear.bias.data = conv.bias.data
         return linear
 
-    def forward(self, cls_feat: torch.Tensor, loc_feat: torch.Tensor, conf: float) -> Tuple[Tuple, torch.Tensor]:
+    def forward(self, cls_feat: torch.Tensor, loc_feat: torch.Tensor, conf: float) -> tuple[tuple, torch.Tensor]:
         """Process classification and localization features to generate detection proposals."""
         if self.enabled:
             pf_score = self.pf(cls_feat)[0, 0].flatten(0)
@@ -625,7 +635,7 @@ class YOLOEDetect(Detect):
 
     is_fused = False
 
-    def __init__(self, nc: int = 80, embed: int = 512, with_bn: bool = False, ch: Tuple = ()):
+    def __init__(self, nc: int = 80, embed: int = 512, with_bn: bool = False, ch: tuple = ()):
         """
         Initialize YOLO detection layer with nc classes and layer channels ch.
 
@@ -638,7 +648,7 @@ class YOLOEDetect(Detect):
         super().__init__(nc, ch)
         c3 = max(ch[0], min(self.nc, 100))
         assert c3 <= embed
-        assert with_bn is True
+        assert with_bn
         self.cv3 = (
             nn.ModuleList(nn.Sequential(Conv(x, c3, 3), Conv(c3, c3, 3), nn.Conv2d(c3, embed, 1)) for x in ch)
             if self.legacy
@@ -705,11 +715,11 @@ class YOLOEDetect(Detect):
         self.reprta = nn.Identity()
         self.is_fused = True
 
-    def get_tpe(self, tpe: Optional[torch.Tensor]) -> Optional[torch.Tensor]:
+    def get_tpe(self, tpe: torch.Tensor | None) -> torch.Tensor | None:
         """Get text prompt embeddings with normalization."""
         return None if tpe is None else F.normalize(self.reprta(tpe), dim=-1, p=2)
 
-    def get_vpe(self, x: List[torch.Tensor], vpe: torch.Tensor) -> torch.Tensor:
+    def get_vpe(self, x: list[torch.Tensor], vpe: torch.Tensor) -> torch.Tensor:
         """Get visual prompt embeddings with spatial awareness."""
         if vpe.shape[1] == 0:  # no visual prompt embeddings
             return torch.zeros(x[0].shape[0], 0, self.embed, device=x[0].device)
@@ -718,7 +728,7 @@ class YOLOEDetect(Detect):
         assert vpe.ndim == 3  # (B, N, D)
         return vpe
 
-    def forward_lrpc(self, x: List[torch.Tensor], return_mask: bool = False) -> Union[torch.Tensor, Tuple]:
+    def forward_lrpc(self, x: list[torch.Tensor], return_mask: bool = False) -> torch.Tensor | tuple:
         """Process features with fused text embeddings to generate detections for prompt-free model."""
         masks = []
         assert self.is_fused, "Prompt-free inference requires model to be fused!"
@@ -756,9 +766,7 @@ class YOLOEDetect(Detect):
         else:
             return y if self.export else (y, x)
 
-    def forward(
-        self, x: List[torch.Tensor], cls_pe: torch.Tensor, return_mask: bool = False
-    ) -> Union[torch.Tensor, Tuple]:
+    def forward(self, x: list[torch.Tensor], cls_pe: torch.Tensor, return_mask: bool = False) -> torch.Tensor | tuple:
         """Process features with class prompt embeddings to generate detections."""
         if hasattr(self, "lrpc"):  # for prompt-free inference
             return self.forward_lrpc(x, return_mask)
@@ -807,7 +815,7 @@ class YOLOESegment(YOLOEDetect):
     """
 
     def __init__(
-        self, nc: int = 80, nm: int = 32, npr: int = 256, embed: int = 512, with_bn: bool = False, ch: Tuple = ()
+        self, nc: int = 80, nm: int = 32, npr: int = 256, embed: int = 512, with_bn: bool = False, ch: tuple = ()
     ):
         """
         Initialize YOLOESegment with class count, mask parameters, and embedding dimensions.
@@ -828,7 +836,7 @@ class YOLOESegment(YOLOEDetect):
         c5 = max(ch[0] // 4, self.nm)
         self.cv5 = nn.ModuleList(nn.Sequential(Conv(x, c5, 3), Conv(c5, c5, 3), nn.Conv2d(c5, self.nm, 1)) for x in ch)
 
-    def forward(self, x: List[torch.Tensor], text: torch.Tensor) -> Union[Tuple, torch.Tensor]:
+    def forward(self, x: list[torch.Tensor], text: torch.Tensor) -> tuple | torch.Tensor:
         """Return model outputs and mask coefficients if training, otherwise return outputs and mask coefficients."""
         p = self.proto(x[0])  # mask protos
         bs = p.shape[0]  # batch size
@@ -896,7 +904,7 @@ class RTDETRDecoder(nn.Module):
     def __init__(
         self,
         nc: int = 80,
-        ch: Tuple = (512, 1024, 2048),
+        ch: tuple = (512, 1024, 2048),
         hd: int = 256,  # hidden dim
         nq: int = 300,  # num queries
         ndp: int = 4,  # num decoder points
@@ -972,7 +980,7 @@ class RTDETRDecoder(nn.Module):
 
         self._reset_parameters()
 
-    def forward(self, x: List[torch.Tensor], batch: Optional[dict] = None) -> Union[Tuple, torch.Tensor]:
+    def forward(self, x: list[torch.Tensor], batch: dict | None = None) -> tuple | torch.Tensor:
         """
         Run the forward pass of the module, returning bounding box and classification scores for the input.
 
@@ -1024,12 +1032,12 @@ class RTDETRDecoder(nn.Module):
 
     def _generate_anchors(
         self,
-        shapes: List[List[int]],
+        shapes: list[list[int]],
         grid_size: float = 0.05,
         dtype: torch.dtype = torch.float32,
         device: str = "cpu",
         eps: float = 1e-2,
-    ) -> Tuple[torch.Tensor, torch.Tensor]:
+    ) -> tuple[torch.Tensor, torch.Tensor]:
         """
         Generate anchor bounding boxes for given shapes with specific grid size and validate them.
 
@@ -1062,7 +1070,7 @@ class RTDETRDecoder(nn.Module):
         anchors = anchors.masked_fill(~valid_mask, float("inf"))
         return anchors, valid_mask
 
-    def _get_encoder_input(self, x: List[torch.Tensor]) -> Tuple[torch.Tensor, List[List[int]]]:
+    def _get_encoder_input(self, x: list[torch.Tensor]) -> tuple[torch.Tensor, list[list[int]]]:
         """
         Process and return encoder inputs by getting projection features from input and concatenating them.
 
@@ -1092,10 +1100,10 @@ class RTDETRDecoder(nn.Module):
     def _get_decoder_input(
         self,
         feats: torch.Tensor,
-        shapes: List[List[int]],
-        dn_embed: Optional[torch.Tensor] = None,
-        dn_bbox: Optional[torch.Tensor] = None,
-    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
+        shapes: list[list[int]],
+        dn_embed: torch.Tensor | None = None,
+        dn_bbox: torch.Tensor | None = None,
+    ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
         """
         Generate and prepare the input required for the decoder from the provided features and shapes.
 
@@ -1200,7 +1208,7 @@ class v10Detect(Detect):
 
     end2end = True
 
-    def __init__(self, nc: int = 80, ch: Tuple = ()):
+    def __init__(self, nc: int = 80, ch: tuple = ()):
         """
         Initialize the v10Detect object with the specified number of classes and input channels.
 

ultralytics/nn/modules/transformer.py

@@ -1,8 +1,9 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 """Transformer modules."""
 
+from __future__ import annotations
+
 import math
-from typing import List, Optional
 
 import torch
 import torch.nn as nn
@@ -88,16 +89,16 @@ class TransformerEncoderLayer(nn.Module):
         self.normalize_before = normalize_before
 
     @staticmethod
-    def with_pos_embed(tensor: torch.Tensor, pos: Optional[torch.Tensor] = None) -> torch.Tensor:
+    def with_pos_embed(tensor: torch.Tensor, pos: torch.Tensor | None = None) -> torch.Tensor:
         """Add position embeddings to the tensor if provided."""
         return tensor if pos is None else tensor + pos
 
     def forward_post(
         self,
         src: torch.Tensor,
-        src_mask: Optional[torch.Tensor] = None,
-        src_key_padding_mask: Optional[torch.Tensor] = None,
-        pos: Optional[torch.Tensor] = None,
+        src_mask: torch.Tensor | None = None,
+        src_key_padding_mask: torch.Tensor | None = None,
+        pos: torch.Tensor | None = None,
     ) -> torch.Tensor:
         """
         Perform forward pass with post-normalization.
@@ -122,9 +123,9 @@ class TransformerEncoderLayer(nn.Module):
     def forward_pre(
         self,
         src: torch.Tensor,
-        src_mask: Optional[torch.Tensor] = None,
-        src_key_padding_mask: Optional[torch.Tensor] = None,
-        pos: Optional[torch.Tensor] = None,
+        src_mask: torch.Tensor | None = None,
+        src_key_padding_mask: torch.Tensor | None = None,
+        pos: torch.Tensor | None = None,
     ) -> torch.Tensor:
         """
         Perform forward pass with pre-normalization.
@@ -149,9 +150,9 @@ class TransformerEncoderLayer(nn.Module):
     def forward(
         self,
         src: torch.Tensor,
-        src_mask: Optional[torch.Tensor] = None,
-        src_key_padding_mask: Optional[torch.Tensor] = None,
-        pos: Optional[torch.Tensor] = None,
+        src_mask: torch.Tensor | None = None,
+        src_key_padding_mask: torch.Tensor | None = None,
+        pos: torch.Tensor | None = None,
     ) -> torch.Tensor:
         """
         Forward propagate the input through the encoder module.
@@ -533,8 +534,8 @@ class MSDeformAttn(nn.Module):
         query: torch.Tensor,
         refer_bbox: torch.Tensor,
         value: torch.Tensor,
-        value_shapes: List,
-        value_mask: Optional[torch.Tensor] = None,
+        value_shapes: list,
+        value_mask: torch.Tensor | None = None,
     ) -> torch.Tensor:
         """
         Perform forward pass for multiscale deformable attention.
@@ -649,7 +650,7 @@ class DeformableTransformerDecoderLayer(nn.Module):
         self.norm3 = nn.LayerNorm(d_model)
 
     @staticmethod
-    def with_pos_embed(tensor: torch.Tensor, pos: Optional[torch.Tensor]) -> torch.Tensor:
+    def with_pos_embed(tensor: torch.Tensor, pos: torch.Tensor | None) -> torch.Tensor:
         """Add positional embeddings to the input tensor, if provided."""
         return tensor if pos is None else tensor + pos
 
@@ -672,10 +673,10 @@ class DeformableTransformerDecoderLayer(nn.Module):
         embed: torch.Tensor,
         refer_bbox: torch.Tensor,
         feats: torch.Tensor,
-        shapes: List,
-        padding_mask: Optional[torch.Tensor] = None,
-        attn_mask: Optional[torch.Tensor] = None,
-        query_pos: Optional[torch.Tensor] = None,
+        shapes: list,
+        padding_mask: torch.Tensor | None = None,
+        attn_mask: torch.Tensor | None = None,
+        query_pos: torch.Tensor | None = None,
     ) -> torch.Tensor:
         """
         Perform the forward pass through the entire decoder layer.
@@ -749,12 +750,12 @@ class DeformableTransformerDecoder(nn.Module):
         embed: torch.Tensor,  # decoder embeddings
         refer_bbox: torch.Tensor,  # anchor
         feats: torch.Tensor,  # image features
-        shapes: List,  # feature shapes
+        shapes: list,  # feature shapes
         bbox_head: nn.Module,
         score_head: nn.Module,
         pos_mlp: nn.Module,
-        attn_mask: Optional[torch.Tensor] = None,
-        padding_mask: Optional[torch.Tensor] = None,
+        attn_mask: torch.Tensor | None = None,
+        padding_mask: torch.Tensor | None = None,
     ):
         """
         Perform the forward pass through the entire decoder.

ultralytics/nn/tasks.py

@@ -1548,7 +1548,7 @@ def attempt_load_one_weight(weight, device=None, inplace=True, fuse=False):
     """
     ckpt, weight = torch_safe_load(weight)  # load ckpt
     args = {**DEFAULT_CFG_DICT, **(ckpt.get("train_args", {}))}  # combine model and default args, preferring model args
-    model = (ckpt.get("ema") or ckpt["model"]).to(device).float()  # FP32 model
+    model = (ckpt.get("ema") or ckpt["model"]).float()  # FP32 model
 
     # Model compatibility updates
     model.args = {k: v for k, v in args.items() if k in DEFAULT_CFG_KEYS}  # attach args to model
@@ -1557,7 +1557,7 @@ def attempt_load_one_weight(weight, device=None, inplace=True, fuse=False):
     if not hasattr(model, "stride"):
         model.stride = torch.tensor([32.0])
 
-    model = model.fuse().eval() if fuse and hasattr(model, "fuse") else model.eval()  # model in eval mode
+    model = (model.fuse() if fuse and hasattr(model, "fuse") else model).eval().to(device)  # model in eval mode
 
     # Module updates
     for m in model.modules():

ultralytics/nn/text_model.py

@@ -1,8 +1,9 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 
+from __future__ import annotations
+
 from abc import abstractmethod
 from pathlib import Path
-from typing import List, Union
 
 import torch
 import torch.nn as nn
@@ -91,7 +92,7 @@ class CLIP(TextModel):
         self.device = device
         self.eval()
 
-    def tokenize(self, texts: Union[str, List[str]]) -> torch.Tensor:
+    def tokenize(self, texts: str | list[str]) -> torch.Tensor:
         """
         Convert input texts to CLIP tokens.
 
@@ -135,7 +136,7 @@ class CLIP(TextModel):
         return txt_feats
 
     @smart_inference_mode()
-    def encode_image(self, image: Union[Image.Image, torch.Tensor], dtype: torch.dtype = torch.float32) -> torch.Tensor:
+    def encode_image(self, image: Image.Image | torch.Tensor, dtype: torch.dtype = torch.float32) -> torch.Tensor:
         """
         Encode preprocessed images into normalized feature vectors.
 
@@ -234,7 +235,7 @@ class MobileCLIP(TextModel):
         self.device = device
         self.eval()
 
-    def tokenize(self, texts: List[str]) -> torch.Tensor:
+    def tokenize(self, texts: list[str]) -> torch.Tensor:
         """
         Convert input texts to MobileCLIP tokens.
 
@@ -319,7 +320,7 @@ class MobileCLIPTS(TextModel):
         self.tokenizer = clip.clip.tokenize
         self.device = device
 
-    def tokenize(self, texts: List[str]) -> torch.Tensor:
+    def tokenize(self, texts: list[str]) -> torch.Tensor:
         """
         Convert input texts to MobileCLIP tokens.
 

ultralytics/solutions/analytics.py

@@ -1,7 +1,9 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 
+from __future__ import annotations
+
 from itertools import cycle
-from typing import Any, Dict, Optional
+from typing import Any
 
 import cv2
 import numpy as np
@@ -135,7 +137,7 @@ class Analytics(BaseSolution):
         return SolutionResults(plot_im=plot_im, total_tracks=len(self.track_ids), classwise_count=self.clswise_count)
 
     def update_graph(
-        self, frame_number: int, count_dict: Optional[Dict[str, int]] = None, plot: str = "line"
+        self, frame_number: int, count_dict: dict[str, int] | None = None, plot: str = "line"
     ) -> np.ndarray:
         """
         Update the graph with new data for single or multiple classes.
@@ -204,7 +206,7 @@ class Analytics(BaseSolution):
                         markersize=self.line_width * 5,
                         label=f"{key} Data Points",
                     )
-            if plot == "bar":
+            elif plot == "bar":
                 self.ax.clear()  # clear bar data
                 for label in labels:  # Map labels to colors
                     if label not in self.color_mapping:
@@ -224,12 +226,12 @@ class Analytics(BaseSolution):
                 for bar, label in zip(bars, labels):
                     bar.set_label(label)  # Assign label to each bar
                 self.ax.legend(loc="upper left", fontsize=13, facecolor=self.fg_color, edgecolor=self.fg_color)
-            if plot == "pie":
+            elif plot == "pie":
                 total = sum(counts)
                 percentages = [size / total * 100 for size in counts]
-                start_angle = 90
                 self.ax.clear()
 
+                start_angle = 90
                 # Create pie chart and create legend labels with percentages
                 wedges, _ = self.ax.pie(
                     counts, labels=labels, startangle=start_angle, textprops={"color": self.fg_color}, autopct=None