ultralytics 8.2.69__py3-none-any.whl → 8.2.71__py3-none-any.whl

ultralytics/models/sam2/modules/utils.py

@@ -0,0 +1,191 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
+import torch
+import torch.nn.functional as F
+
+
+def select_closest_cond_frames(frame_idx, cond_frame_outputs, max_cond_frame_num):
+    """
+    Selects the closest conditioning frames to a given frame index.
+
+    Args:
+        frame_idx (int): Current frame index.
+        cond_frame_outputs (Dict[int, Any]): Dictionary of conditioning frame outputs keyed by frame indices.
+        max_cond_frame_num (int): Maximum number of conditioning frames to select.
+
+    Returns:
+        (Tuple[Dict[int, Any], Dict[int, Any]]): A tuple containing two dictionaries:
+            - selected_outputs: Selected items from cond_frame_outputs.
+            - unselected_outputs: Items not selected from cond_frame_outputs.
+
+    Examples:
+        >>> frame_idx = 5
+        >>> cond_frame_outputs = {1: 'a', 3: 'b', 7: 'c', 9: 'd'}
+        >>> max_cond_frame_num = 2
+        >>> selected, unselected = select_closest_cond_frames(frame_idx, cond_frame_outputs, max_cond_frame_num)
+        >>> print(selected)
+        {3: 'b', 7: 'c'}
+        >>> print(unselected)
+        {1: 'a', 9: 'd'}
+    """
+    if max_cond_frame_num == -1 or len(cond_frame_outputs) <= max_cond_frame_num:
+        selected_outputs = cond_frame_outputs
+        unselected_outputs = {}
+    else:
+        assert max_cond_frame_num >= 2, "we should allow using 2+ conditioning frames"
+        selected_outputs = {}
+
+        # the closest conditioning frame before `frame_idx` (if any)
+        idx_before = max((t for t in cond_frame_outputs if t < frame_idx), default=None)
+        if idx_before is not None:
+            selected_outputs[idx_before] = cond_frame_outputs[idx_before]
+
+        # the closest conditioning frame after `frame_idx` (if any)
+        idx_after = min((t for t in cond_frame_outputs if t >= frame_idx), default=None)
+        if idx_after is not None:
+            selected_outputs[idx_after] = cond_frame_outputs[idx_after]
+
+        # add other temporally closest conditioning frames until reaching a total
+        # of `max_cond_frame_num` conditioning frames.
+        num_remain = max_cond_frame_num - len(selected_outputs)
+        inds_remain = sorted(
+            (t for t in cond_frame_outputs if t not in selected_outputs),
+            key=lambda x: abs(x - frame_idx),
+        )[:num_remain]
+        selected_outputs.update((t, cond_frame_outputs[t]) for t in inds_remain)
+        unselected_outputs = {t: v for t, v in cond_frame_outputs.items() if t not in selected_outputs}
+
+    return selected_outputs, unselected_outputs
+
+
+def get_1d_sine_pe(pos_inds, dim, temperature=10000):
+    """Generates 1D sinusoidal positional embeddings for given positions and dimensions."""
+    pe_dim = dim // 2
+    dim_t = torch.arange(pe_dim, dtype=torch.float32, device=pos_inds.device)
+    dim_t = temperature ** (2 * (dim_t // 2) / pe_dim)
+
+    pos_embed = pos_inds.unsqueeze(-1) / dim_t
+    pos_embed = torch.cat([pos_embed.sin(), pos_embed.cos()], dim=-1)
+    return pos_embed
+
+
+def init_t_xy(end_x: int, end_y: int):
+    """Initializes 1D and 2D coordinate tensors for a grid of size end_x by end_y."""
+    t = torch.arange(end_x * end_y, dtype=torch.float32)
+    t_x = (t % end_x).float()
+    t_y = torch.div(t, end_x, rounding_mode="floor").float()
+    return t_x, t_y
+
+
+def compute_axial_cis(dim: int, end_x: int, end_y: int, theta: float = 10000.0):
+    """Computes axial complex exponential positional encodings for 2D spatial positions."""
+    freqs_x = 1.0 / (theta ** (torch.arange(0, dim, 4)[: (dim // 4)].float() / dim))
+    freqs_y = 1.0 / (theta ** (torch.arange(0, dim, 4)[: (dim // 4)].float() / dim))
+
+    t_x, t_y = init_t_xy(end_x, end_y)
+    freqs_x = torch.outer(t_x, freqs_x)
+    freqs_y = torch.outer(t_y, freqs_y)
+    freqs_cis_x = torch.polar(torch.ones_like(freqs_x), freqs_x)
+    freqs_cis_y = torch.polar(torch.ones_like(freqs_y), freqs_y)
+    return torch.cat([freqs_cis_x, freqs_cis_y], dim=-1)
+
+
+def reshape_for_broadcast(freqs_cis: torch.Tensor, x: torch.Tensor):
+    """Reshapes frequency tensor for broadcasting, ensuring compatibility with input tensor dimensions."""
+    ndim = x.ndim
+    assert 0 <= 1 < ndim
+    assert freqs_cis.shape == (x.shape[-2], x.shape[-1])
+    shape = [d if i >= ndim - 2 else 1 for i, d in enumerate(x.shape)]
+    return freqs_cis.view(*shape)
+
+
+def apply_rotary_enc(
+    xq: torch.Tensor,
+    xk: torch.Tensor,
+    freqs_cis: torch.Tensor,
+    repeat_freqs_k: bool = False,
+):
+    """Applies rotary positional encoding to query and key tensors using complex-valued frequency components."""
+    xq_ = torch.view_as_complex(xq.float().reshape(*xq.shape[:-1], -1, 2))
+    xk_ = torch.view_as_complex(xk.float().reshape(*xk.shape[:-1], -1, 2)) if xk.shape[-2] != 0 else None
+    freqs_cis = reshape_for_broadcast(freqs_cis, xq_)
+    xq_out = torch.view_as_real(xq_ * freqs_cis).flatten(3)
+    if xk_ is None:
+        # no keys to rotate, due to dropout
+        return xq_out.type_as(xq).to(xq.device), xk
+    # repeat freqs along seq_len dim to match k seq_len
+    if repeat_freqs_k:
+        r = xk_.shape[-2] // xq_.shape[-2]
+        freqs_cis = freqs_cis.repeat(*([1] * (freqs_cis.ndim - 2)), r, 1)
+    xk_out = torch.view_as_real(xk_ * freqs_cis).flatten(3)
+    return xq_out.type_as(xq).to(xq.device), xk_out.type_as(xk).to(xk.device)
+
+
+def window_partition(x, window_size):
+    """
+    Partitions input tensor into non-overlapping windows with padding if needed.
+
+    Args:
+        x (torch.Tensor): Input tensor with shape (B, H, W, C).
+        window_size (int): Size of each window.
+
+    Returns:
+        (Tuple[torch.Tensor, Tuple[int, int]]): A tuple containing:
+            - windows (torch.Tensor): Partitioned windows with shape (B * num_windows, window_size, window_size, C).
+            - (Hp, Wp) (Tuple[int, int]): Padded height and width before partition.
+
+    Examples:
+        >>> x = torch.randn(1, 16, 16, 3)
+        >>> windows, (Hp, Wp) = window_partition(x, window_size=4)
+        >>> print(windows.shape, Hp, Wp)
+        torch.Size([16, 4, 4, 3]) 16 16
+    """
+    B, H, W, C = x.shape
+
+    pad_h = (window_size - H % window_size) % window_size
+    pad_w = (window_size - W % window_size) % window_size
+    if pad_h > 0 or pad_w > 0:
+        x = F.pad(x, (0, 0, 0, pad_w, 0, pad_h))
+    Hp, Wp = H + pad_h, W + pad_w
+
+    x = x.view(B, Hp // window_size, window_size, Wp // window_size, window_size, C)
+    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
+    return windows, (Hp, Wp)
+
+
+def window_unpartition(windows, window_size, pad_hw, hw):
+    """
+    Unpartitions windowed sequences into original sequences and removes padding.
+
+    This function reverses the windowing process, reconstructing the original input from windowed segments
+    and removing any padding that was added during the windowing process.
+
+    Args:
+        windows (torch.Tensor): Input tensor of windowed sequences with shape (B * num_windows, window_size,
+            window_size, C), where B is the batch size, num_windows is the number of windows, window_size is
+            the size of each window, and C is the number of channels.
+        window_size (int): Size of each window.
+        pad_hw (Tuple[int, int]): Padded height and width (Hp, Wp) of the input before windowing.
+        hw (Tuple[int, int]): Original height and width (H, W) of the input before padding and windowing.
+
+    Returns:
+        (torch.Tensor): Unpartitioned sequences with shape (B, H, W, C), where B is the batch size, H and W
+            are the original height and width, and C is the number of channels.
+
+    Examples:
+        >>> windows = torch.rand(32, 8, 8, 64)  # 32 windows of size 8x8 with 64 channels
+        >>> pad_hw = (16, 16)  # Padded height and width
+        >>> hw = (15, 14)  # Original height and width
+        >>> x = window_unpartition(windows, window_size=8, pad_hw=pad_hw, hw=hw)
+        >>> print(x.shape)
+        torch.Size([1, 15, 14, 64])
+    """
+    Hp, Wp = pad_hw
+    H, W = hw
+    B = windows.shape[0] // (Hp * Wp // window_size // window_size)
+    x = windows.view(B, Hp // window_size, Wp // window_size, window_size, window_size, -1)
+    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, Hp, Wp, -1)
+
+    if Hp > H or Wp > W:
+        x = x[:, :H, :W, :].contiguous()
+    return x

ultralytics/models/sam2/predict.py

@@ -0,0 +1,182 @@
+# Ultralytics YOLO 🚀, AGPL-3.0 license
+
+import torch
+
+from ..sam.predict import Predictor
+from .build import build_sam2
+
+
+class SAM2Predictor(Predictor):
+    """
+    A predictor class for the Segment Anything Model 2 (SAM2), extending the base Predictor class.
+
+    This class provides an interface for model inference tailored to image segmentation tasks, leveraging SAM2's
+    advanced architecture and promptable segmentation capabilities. It facilitates flexible and real-time mask
+    generation, working with various types of prompts such as bounding boxes, points, and low-resolution masks.
+
+    Attributes:
+        cfg (Dict): Configuration dictionary specifying model and task-related parameters.
+        overrides (Dict): Dictionary containing values that override the default configuration.
+        _callbacks (Dict): Dictionary of user-defined callback functions to augment behavior.
+        args (namespace): Namespace to hold command-line arguments or other operational variables.
+        im (torch.Tensor): Preprocessed input image tensor.
+        features (torch.Tensor): Extracted image features used for inference.
+        prompts (Dict): Collection of various prompt types, such as bounding boxes and points.
+        segment_all (bool): Flag to control whether to segment all objects in the image or only specified ones.
+        model (torch.nn.Module): The loaded SAM2 model.
+        device (torch.device): The device (CPU or GPU) on which the model is loaded.
+        _bb_feat_sizes (List[Tuple[int, int]]): List of feature sizes for different backbone levels.
+
+    Methods:
+        get_model: Builds and returns the SAM2 model.
+        prompt_inference: Performs image segmentation inference based on various prompts.
+        set_image: Preprocesses and sets a single image for inference.
+        get_im_features: Extracts image features from the SAM2 image encoder.
+
+    Examples:
+        >>> predictor = SAM2Predictor(model='sam2_l.pt')
+        >>> predictor.set_image('path/to/image.jpg')
+        >>> masks, scores = predictor.prompt_inference(im=predictor.im, points=[[500, 375]], labels=[1])
+        >>> print(f"Generated {len(masks)} mask(s) with scores: {scores}")
+    """
+
+    _bb_feat_sizes = [
+        (256, 256),
+        (128, 128),
+        (64, 64),
+    ]
+
+    def get_model(self):
+        """Retrieves and initializes the Segment Anything Model (SAM) for image segmentation tasks."""
+        return build_sam2(self.args.model)
+
+    def prompt_inference(
+        self,
+        im,
+        bboxes=None,
+        points=None,
+        labels=None,
+        masks=None,
+        multimask_output=False,
+        img_idx=-1,
+    ):
+        """
+        Performs image segmentation inference based on various prompts using SAM2 architecture.
+
+        Args:
+            im (torch.Tensor): Preprocessed input image tensor with shape (N, C, H, W).
+            bboxes (np.ndarray | List | None): Bounding boxes in XYXY format with shape (N, 4).
+            points (np.ndarray | List | None): Points indicating object locations with shape (N, 2), in pixels.
+            labels (np.ndarray | List | None): Labels for point prompts with shape (N,). 1 = foreground, 0 = background.
+            masks (np.ndarray | None): Low-resolution masks from previous predictions with shape (N, H, W).
+            multimask_output (bool): Flag to return multiple masks for ambiguous prompts.
+            img_idx (int): Index of the image in the batch to process.
+
+        Returns:
+            (tuple): Tuple containing:
+                - np.ndarray: Output masks with shape (C, H, W), where C is the number of generated masks.
+                - np.ndarray: Quality scores for each mask, with length C.
+                - np.ndarray: Low-resolution logits with shape (C, 256, 256) for subsequent inference.
+
+        Examples:
+            >>> predictor = SAM2Predictor(cfg)
+            >>> image = torch.rand(1, 3, 640, 640)
+            >>> bboxes = [[100, 100, 200, 200]]
+            >>> masks, scores, logits = predictor.prompt_inference(image, bboxes=bboxes)
+        """
+        features = self.get_im_features(im) if self.features is None else self.features
+
+        src_shape, dst_shape = self.batch[1][0].shape[:2], im.shape[2:]
+        r = 1.0 if self.segment_all else min(dst_shape[0] / src_shape[0], dst_shape[1] / src_shape[1])
+        # Transform input prompts
+        if points is not None:
+            points = torch.as_tensor(points, dtype=torch.float32, device=self.device)
+            points = points[None] if points.ndim == 1 else points
+            # Assuming labels are all positive if users don't pass labels.
+            if labels is None:
+                labels = torch.ones(points.shape[0])
+            labels = torch.as_tensor(labels, dtype=torch.int32, device=self.device)
+            points *= r
+            # (N, 2) --> (N, 1, 2), (N, ) --> (N, 1)
+            points, labels = points[:, None], labels[:, None]
+        if bboxes is not None:
+            bboxes = torch.as_tensor(bboxes, dtype=torch.float32, device=self.device)
+            bboxes = bboxes[None] if bboxes.ndim == 1 else bboxes
+            bboxes *= r
+        if masks is not None:
+            masks = torch.as_tensor(masks, dtype=torch.float32, device=self.device).unsqueeze(1)
+
+        points = (points, labels) if points is not None else None
+        # TODO: Embed prompts
+        # if bboxes is not None:
+        #     box_coords = bboxes.reshape(-1, 2, 2)
+        #     box_labels = torch.tensor([[2, 3]], dtype=torch.int, device=bboxes.device)
+        #     box_labels = box_labels.repeat(bboxes.size(0), 1)
+        #     # we merge "boxes" and "points" into a single "concat_points" input (where
+        #     # boxes are added at the beginning) to sam_prompt_encoder
+        #     if concat_points is not None:
+        #         concat_coords = torch.cat([box_coords, concat_points[0]], dim=1)
+        #         concat_labels = torch.cat([box_labels, concat_points[1]], dim=1)
+        #         concat_points = (concat_coords, concat_labels)
+        #     else:
+        #         concat_points = (box_coords, box_labels)
+
+        sparse_embeddings, dense_embeddings = self.model.sam_prompt_encoder(
+            points=points,
+            boxes=bboxes,
+            masks=masks,
+        )
+        # Predict masks
+        batched_mode = points is not None and points[0].shape[0] > 1  # multi object prediction
+        high_res_features = [feat_level[img_idx].unsqueeze(0) for feat_level in features["high_res_feats"]]
+        pred_masks, pred_scores, _, _ = self.model.sam_mask_decoder(
+            image_embeddings=features["image_embed"][img_idx].unsqueeze(0),
+            image_pe=self.model.sam_prompt_encoder.get_dense_pe(),
+            sparse_prompt_embeddings=sparse_embeddings,
+            dense_prompt_embeddings=dense_embeddings,
+            multimask_output=multimask_output,
+            repeat_image=batched_mode,
+            high_res_features=high_res_features,
+        )
+        # (N, d, H, W) --> (N*d, H, W), (N, d) --> (N*d, )
+        # `d` could be 1 or 3 depends on `multimask_output`.
+        return pred_masks.flatten(0, 1), pred_scores.flatten(0, 1)
+
+    def set_image(self, image):
+        """
+        Preprocesses and sets a single image for inference.
+
+        This function sets up the model if not already initialized, configures the data source to the specified image,
+        and preprocesses the image for feature extraction. Only one image can be set at a time.
+
+        Args:
+            image (str | np.ndarray): Image file path as a string, or a numpy array image read by cv2.
+
+        Raises:
+            AssertionError: If more than one image is set.
+
+        Examples:
+            >>> predictor = SAM2Predictor()
+            >>> predictor.set_image("path/to/image.jpg")
+            >>> predictor.set_image(np.array([...]))  # Using a numpy array
+        """
+        if self.model is None:
+            self.setup_model(model=None)
+        self.setup_source(image)
+        assert len(self.dataset) == 1, "`set_image` only supports setting one image!"
+        for batch in self.dataset:
+            im = self.preprocess(batch[1])
+            self.features = self.get_im_features(im)
+            break
+
+    def get_im_features(self, im):
+        """Extracts and processes image features using SAM2's image encoder for subsequent segmentation tasks."""
+        backbone_out = self.model.forward_image(im)
+        _, vision_feats, _, _ = self.model._prepare_backbone_features(backbone_out)
+        if self.model.directly_add_no_mem_embed:
+            vision_feats[-1] = vision_feats[-1] + self.model.no_mem_embed
+        feats = [
+            feat.permute(1, 2, 0).view(1, -1, *feat_size)
+            for feat, feat_size in zip(vision_feats[::-1], self._bb_feat_sizes[::-1])
+        ][::-1]
+        return {"image_embed": feats[-1], "high_res_feats": feats[:-1]}

ultralytics/nn/modules/transformer.py

@@ -174,18 +174,20 @@ class MLPBlock(nn.Module):
 class MLP(nn.Module):
     """Implements a simple multi-layer perceptron (also called FFN)."""
 
-    def __init__(self, input_dim, hidden_dim, output_dim, num_layers):
+    def __init__(self, input_dim, hidden_dim, output_dim, num_layers, act=nn.ReLU, sigmoid=False):
         """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]))
+        self.sigmoid = sigmoid
+        self.act = act()
 
     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
+            x = self.act(layer(x)) if i < self.num_layers - 1 else layer(x)
+        return x.sigmoid() if self.sigmoid else x
 
 
 class LayerNorm2d(nn.Module):

ultralytics/utils/__init__.py

@@ -47,7 +47,7 @@ PYTHON_VERSION = platform.python_version()
 TORCH_VERSION = torch.__version__
 TORCHVISION_VERSION = importlib.metadata.version("torchvision")  # faster than importing torchvision
 HELP_MSG = """
-    Usage examples for running YOLOv8:
+    Usage examples for running Ultralytics YOLO:
 
     1. Install the ultralytics package:
 
@@ -58,25 +58,25 @@ HELP_MSG = """
         from ultralytics import YOLO
 
         # Load a model
-        model = YOLO('yolov8n.yaml')  # build a new model from scratch
+        model = YOLO("yolov8n.yaml")  # build a new model from scratch
         model = YOLO("yolov8n.pt")  # load a pretrained model (recommended for training)
 
         # Use the model
         results = model.train(data="coco8.yaml", epochs=3)  # train the model
         results = model.val()  # evaluate model performance on the validation set
-        results = model('https://ultralytics.com/images/bus.jpg')  # predict on an image
-        success = model.export(format='onnx')  # export the model to ONNX format
+        results = model("https://ultralytics.com/images/bus.jpg")  # predict on an image
+        success = model.export(format="onnx")  # export the model to ONNX format
 
     3. Use the command line interface (CLI):
 
-        YOLOv8 'yolo' CLI commands use the following syntax:
+        Ultralytics 'yolo' CLI commands use the following syntax:
 
             yolo TASK MODE ARGS
 
-            Where   TASK (optional) is one of [detect, segment, classify]
-                    MODE (required) is one of [train, val, predict, export]
-                    ARGS (optional) are any number of custom 'arg=value' pairs like 'imgsz=320' that override defaults.
-                        See all ARGS at https://docs.ultralytics.com/usage/cfg or with 'yolo cfg'
+            Where   TASK (optional) is one of [detect, segment, classify, pose, obb]
+                    MODE (required) is one of [train, val, predict, export, benchmark]
+                    ARGS (optional) are any number of custom "arg=value" pairs like "imgsz=320" that override defaults.
+                        See all ARGS at https://docs.ultralytics.com/usage/cfg or with "yolo cfg"
 
         - Train a detection model for 10 epochs with an initial learning_rate of 0.01
             yolo detect train data=coco8.yaml model=yolov8n.pt epochs=10 lr0=0.01

ultralytics/utils/plotting.py

@@ -280,7 +280,7 @@ class Annotator:
         Args:
             box (tuple): The bounding box coordinates (x1, y1, x2, y2).
             label (str): The text label to be displayed.
-            color (tuple, optional): The background color of the rectangle (R, G, B).
+            color (tuple, optional): The background color of the rectangle (B, G, R).
             txt_color (tuple, optional): The color of the text (R, G, B).
             rotated (bool, optional): Variable used to check if task is OBB
         """

ultralytics/utils/torch_utils.py

@@ -1,5 +1,5 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
-
+import contextlib
 import gc
 import math
 import os
@@ -48,11 +48,12 @@ TORCHVISION_0_18 = check_version(TORCHVISION_VERSION, "0.18.0")
 def torch_distributed_zero_first(local_rank: int):
     """Ensures all processes in distributed training wait for the local master (rank 0) to complete a task first."""
     initialized = dist.is_available() and dist.is_initialized()
+
     if initialized and local_rank not in {-1, 0}:
         dist.barrier(device_ids=[local_rank])
     yield
     if initialized and local_rank == 0:
-        dist.barrier(device_ids=[0])
+        dist.barrier(device_ids=[local_rank])
 
 
 def smart_inference_mode():
@@ -101,12 +102,15 @@ def autocast(enabled: bool, device: str = "cuda"):
 
 def get_cpu_info():
     """Return a string with system CPU information, i.e. 'Apple M2'."""
-    import cpuinfo  # pip install py-cpuinfo
+    with contextlib.suppress(Exception):
+        import cpuinfo  # pip install py-cpuinfo
+
+        k = "brand_raw", "hardware_raw", "arch_string_raw"  # keys sorted by preference (not all keys always available)
+        info = cpuinfo.get_cpu_info()  # info dict
+        string = info.get(k[0] if k[0] in info else k[1] if k[1] in info else k[2], "unknown")
+        return string.replace("(R)", "").replace("CPU ", "").replace("@ ", "")
 
-    k = "brand_raw", "hardware_raw", "arch_string_raw"  # info keys sorted by preference (not all keys always available)
-    info = cpuinfo.get_cpu_info()  # info dict
-    string = info.get(k[0] if k[0] in info else k[1] if k[1] in info else k[2], "unknown")
-    return string.replace("(R)", "").replace("CPU ", "").replace("@ ", "")
+    return "unknown"
 
 
 def select_device(device="", batch=0, newline=False, verbose=True):

{ultralytics-8.2.69.dist-info → ultralytics-8.2.71.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ultralytics
-Version: 8.2.69
+Version: 8.2.71
 Summary: Ultralytics YOLOv8 for SOTA object detection, multi-object tracking, instance segmentation, pose estimation and image classification.
 Author: Glenn Jocher, Ayush Chaurasia, Jing Qiu
 Maintainer: Glenn Jocher, Ayush Chaurasia, Jing Qiu

{ultralytics-8.2.69.dist-info → ultralytics-8.2.71.dist-info}/RECORD

@@ -8,10 +8,10 @@ tests/test_exports.py,sha256=Uezf3OatpPHlo5qoPw-2kqkZxuMCF9L4XF2riD4vmII,8225
 tests/test_integrations.py,sha256=xglcfMPjfVh346PV8WTpk6tBxraCXEFJEQyyJMr5tyU,6064
 tests/test_python.py,sha256=cLK8dyRf_4H_znFIm-krnOFMydwkxKlVZvHwl9vbck8,21780
 tests/test_solutions.py,sha256=EACnPXbeJe2aVTOKfqMk5jclKKCWCVgFEzjpR6y7Sh8,3304
-ultralytics/__init__.py,sha256=YWRj4FNGuxXRahBpsPRAOxm3h0rYMQTFTPqJcnwUEDE,694
+ultralytics/__init__.py,sha256=sV3uzVV5yg9sqZSe7JS8hxO873wlJUProzRIeVii45U,712
 ultralytics/assets/bus.jpg,sha256=wCAZxJecGR63Od3ZRERe9Aja1Weayrb9Ug751DS_vGM,137419
 ultralytics/assets/zidane.jpg,sha256=Ftc4aeMmen1O0A3o6GCDO9FlfBslLpTAw0gnetx7bts,50427
-ultralytics/cfg/__init__.py,sha256=fD3Llw12sIkJo4g667t6b051je9nEpwdBLGgbbVEzHY,32973
+ultralytics/cfg/__init__.py,sha256=7ce3_bhi7pDw5ZAbSqYR6e3_IYD2JCLCy7fkl5d1WyI,33064
 ultralytics/cfg/default.yaml,sha256=xRKVF-Z9E3imXTU9OCK94kj3jGgYoo67VJQwuYlHiUU,8228
 ultralytics/cfg/datasets/Argoverse.yaml,sha256=FyeuJT5CHq_9d4hlfAf0kpZlnbUMO0S--UJ1yIqcdKk,3134
 ultralytics/cfg/datasets/DOTAv1.5.yaml,sha256=QVfp_Qp-4rukuicaB4qx86NxSHM8Mrzym8l_fIDo8gw,1195
@@ -87,7 +87,7 @@ ultralytics/data/annotator.py,sha256=1Hyu6ubrBL8KmRrt1keGn-K4XTqQdAVyIwTsQiBtzLU
 ultralytics/data/augment.py,sha256=ExU4khJfJ_TeczkJRLNUDscN57SJvAjnm-reouJcxGI,119309
 ultralytics/data/base.py,sha256=C3teLnw97ZTbpJHT9P7yYWosAKocMzgJjRe1rxgfpls,13524
 ultralytics/data/build.py,sha256=AfMmz0sHIYmwry_90tEJFRk_kz0S3SolScVXqYHiT08,7261
-ultralytics/data/converter.py,sha256=7640xKuf7LPeoTwoCvgbIXM5xbzyq72Hu2Rf2lrgjRY,17554
+ultralytics/data/converter.py,sha256=R9zrsClQGxDBg035o63jNQRALALY3XVAytD9xksVU1M,21442
 ultralytics/data/dataset.py,sha256=ZBnO9KPVOJXwKQbN2LlmROIxLEb0mtppVQlrC4sX3oE,22879
 ultralytics/data/loaders.py,sha256=vy71TzKAPqohCp4MDNQpia2CR1LaOxAU5eA14DonJoU,24085
 ultralytics/data/split_dota.py,sha256=fWezt1Bo3jiZ6AyUWdBtTUuvLamPv1t7JD-DirM9gQ8,10142
@@ -102,18 +102,18 @@ ultralytics/engine/exporter.py,sha256=EM35MOPWbIKE2ShJsPzdrEmrjzwZSp9gW-rO8GEFal
 ultralytics/engine/model.py,sha256=8YSxLan1OfV_IynCQjAzaGS4gCWTEbGLfUWnfTDxhsE,52047
 ultralytics/engine/predictor.py,sha256=W58kDCFH2AfoFzpGbos3k8zUEVsLunBuM8sc2B64rPY,17449
 ultralytics/engine/results.py,sha256=oNAzSKdKxxx_5QQd9opzCevvgPhspdY5BkWxoz5bQ8E,69882
-ultralytics/engine/trainer.py,sha256=esQhG3XJUF1vsl49GavnqpL0tvMZIY-SwD_hw1XmWdU,35454
+ultralytics/engine/trainer.py,sha256=BHcwcWgEiN7wBMEkGfO4AVo3O3M0IreVl1OnHXPipuA,35482
 ultralytics/engine/tuner.py,sha256=iZrgMmXSDpfuDu4bdFRflmAsscys2-8W8qAGxSyOVJE,11844
-ultralytics/engine/validator.py,sha256=Y21Uo8_Zto4qjk_YqQk6k7tyfpq_Qk9cfjeXeyDRxs8,14643
+ultralytics/engine/validator.py,sha256=R0qND144EHTOedGDoqI60MxY9RigbynjvGrRYUjjPsU,14682
 ultralytics/hub/__init__.py,sha256=93bqI8x8-MfDYdKkQVduuocUiQj3WGnk1nIk0li08zA,5663
 ultralytics/hub/auth.py,sha256=FID58NE6fh7Op_B45QOpWBw1qoBN0ponL16uvyb2dZ8,5399
 ultralytics/hub/session.py,sha256=UF_aVwyxnbP-OzpzKXGGhi4i6KGWjjhoj5Qsn46dFpE,16257
 ultralytics/hub/utils.py,sha256=tXfM3QbXBcf4Y6StgHI1pktT4OM7Ic9eF3xiBFHGlhY,9721
 ultralytics/hub/google/__init__.py,sha256=qyvvpGP-4NAtrn7GLqfqxP_aWuRP1T0OvJYafWKvL2Q,7512
-ultralytics/models/__init__.py,sha256=TT9iLCL_n9Y80dcUq0Fo-p-GRZCSU2vrWXM3CoMwqqE,265
+ultralytics/models/__init__.py,sha256=AlVStwxv5pMrYaPL8dLhu4sY2c2JgqmuK__RlEDKrEo,296
 ultralytics/models/fastsam/__init__.py,sha256=W0rRSJM3vdxcsneuiN6_ajkUw86k6-opUKdLxVhKOoQ,203
 ultralytics/models/fastsam/model.py,sha256=r5VZj-KLKaqZtEKTZxQik8vQI2N9uOF4xpV_gA-P8h0,2101
-ultralytics/models/fastsam/predict.py,sha256=ej1Z93W73hThBxuHTdb-LB-yElijKnAMxrTUMlXJ8Qs,7262
+ultralytics/models/fastsam/predict.py,sha256=z5j2IMwf4MURuROKeqNXW1WvOSj91UdJa7dLRqN_OFc,7370
 ultralytics/models/fastsam/utils.py,sha256=dCSm6l5yua_PTT5aNvyOvn1Q0h42Ta_NovO7sTbsBxM,715
 ultralytics/models/fastsam/val.py,sha256=ILKmw3U8FYmmQsO9wk9-bJ9Pyp_ZthJM36b61L75s3Y,1967
 ultralytics/models/nas/__init__.py,sha256=d6-WTrYLXvbPs58ebA0-583ODi-VyzXc-t4aGIDQK6M,179
@@ -127,15 +127,26 @@ ultralytics/models/rtdetr/train.py,sha256=20AFYVW9NPxw0-cp-sRdIovWidFL0IIhJRv2oZ
 ultralytics/models/rtdetr/val.py,sha256=4QQArdaGEY8rJsJuvyJ032f8GGVGdV2jURHK2EdMxyk,5566
 ultralytics/models/sam/__init__.py,sha256=9A1iyfPN_ncqq3TMExe_-uPoARjEX3psoHEI1xMG2VE,144
 ultralytics/models/sam/amg.py,sha256=He2c4nIoZ__F_pL18rRl278R8iBjWXBM2Z_vxfuVOkk,7971
-ultralytics/models/sam/build.py,sha256=-i-vj0egQ2idBZUf3Xf-H89QeToM3ky0HTxKP_KEXTs,4944
-ultralytics/models/sam/model.py,sha256=dkEhqJEZFuSoKubMaAjUx1U9Np49AII3nBScdH8rMBI,4707
-ultralytics/models/sam/predict.py,sha256=hachjdcJ175v_oOUPmu_jG_VSe2wCbpLpi4qymUJV34,23575
+ultralytics/models/sam/build.py,sha256=BSpRgDIQb-kgxaQtSq0C7Zb2UsqkfkFRmErC_bzKYIg,4954
+ultralytics/models/sam/model.py,sha256=MVO7WqF41Sq1-qbsN8O8Fophe8anYVY67yp17Sudp0k,4979
+ultralytics/models/sam/predict.py,sha256=GoR8xCwt3VJdBX5wUUhJ-3qY87LoG-R1eG8SziTDLP0,23755
 ultralytics/models/sam/modules/__init__.py,sha256=mHtJuK4hwF8cuV-VHDc7tp6u6D1gHz2Z7JI8grmQDTs,42
-ultralytics/models/sam/modules/decoders.py,sha256=7NWnBNupxGYvH0S1N0R6NBHxdVFRUrrnL9EqAw09J4E,7816
-ultralytics/models/sam/modules/encoders.py,sha256=pRNZHzt2J2xD_D0Btu8pk4DcItfr6dRr9rcRfxoZZhU,24746
-ultralytics/models/sam/modules/sam.py,sha256=zC4l4kcrIQD_ekczjl2l6dgaABqqjROZxQ-FDb-itt0,2783
+ultralytics/models/sam/modules/decoders.py,sha256=SkVdfsFwy8g4rOJYXi2rWg2zI5HEttsQJi6E4Uwxs9o,6307
+ultralytics/models/sam/modules/encoders.py,sha256=JHot5dRyr_d3wAzz7jgc_SOWOWibWKRWOt-IGBXPhW8,24894
+ultralytics/models/sam/modules/sam.py,sha256=lyB-edOBr85gACTaVqG0WiSIS4FyohTtLqkNMKDwVM0,2695
 ultralytics/models/sam/modules/tiny_encoder.py,sha256=rAY9JuyxUpFivFUUPVjK2aUYlsXEZ0JGKVoEWDGf0Eo,29228
-ultralytics/models/sam/modules/transformer.py,sha256=VINZMb4xkx4IHAbJdhCq2XLDvaFBMup7RGC16DLS7OY,11164
+ultralytics/models/sam/modules/transformer.py,sha256=a2jsS_J76MvrIKIERb_0flliYFMjpBbwVL4UnsNnoyE,11232
+ultralytics/models/sam2/__init__.py,sha256=_xqQHLZTLgEdK278YETYR-Fts2hsvXP5q9ddUbuuFvc,154
+ultralytics/models/sam2/build.py,sha256=m6hv82VKn3Lct_7nztUqdzJzCV9Nbr5mvqpI8nkReQM,5422
+ultralytics/models/sam2/model.py,sha256=PS-eV78DVNrGZmUq7L7gJHgrGjxnySM1TTHkwfrQM7E,3408
+ultralytics/models/sam2/predict.py,sha256=gvKf6qcStFiT9SLzo8Ol25suIh-QRVcOcdbyeuM2ORw,8894
+ultralytics/models/sam2/modules/__init__.py,sha256=mHtJuK4hwF8cuV-VHDc7tp6u6D1gHz2Z7JI8grmQDTs,42
+ultralytics/models/sam2/modules/decoders.py,sha256=t4SR-0g3HQstk-agiapCsVYTMZBFc2vz24zfgBwZUkw,15376
+ultralytics/models/sam2/modules/encoders.py,sha256=0VRK2wdl0vZzKA3528_j-Vyn4Iy8XlNHp2ftQRn-aGE,13313
+ultralytics/models/sam2/modules/memory_attention.py,sha256=4zdvm8_ANM0r8QSN_xBGi9l-9Ugjt3gxBsHv2cHczjc,6214
+ultralytics/models/sam2/modules/sam2.py,sha256=CgCBrfjhKDHI2n8iM6AIJmXeCEgf2_qUz7rzZT31fB0,44255
+ultralytics/models/sam2/modules/sam2_blocks.py,sha256=7HmuZTFw8VVdAVDsIStWByxyUHBqytnfgvQMaCNr1GU,28379
+ultralytics/models/sam2/modules/utils.py,sha256=2H5C3sjBnYoPuoJqflH3AmGeBJoKrhHea136jgwIq_I,8320
 ultralytics/models/utils/__init__.py,sha256=mHtJuK4hwF8cuV-VHDc7tp6u6D1gHz2Z7JI8grmQDTs,42
 ultralytics/models/utils/loss.py,sha256=PmlKDe4xQTiYkPSCdNUabxJC7bh43zGxiKVIxsXBVGE,15135
 ultralytics/models/utils/ops.py,sha256=sAeD_koytXDzHibIvQLLAx3vOpGdhdAiQhMiNFUnn5U,13255
@@ -172,7 +183,7 @@ ultralytics/nn/modules/activation.py,sha256=RS0DRDm9r56tojN79X8UBVtiktde9Wasw7GI
 ultralytics/nn/modules/block.py,sha256=jLXQerl4nXfr4MEGMp9S3YgdTqOJzas1GBxryyXyLV0,34582
 ultralytics/nn/modules/conv.py,sha256=Ywe87IhuaS22mR2JJ9xjnW8Sb-m7WTjxuqIxV_Dv8lI,12722
 ultralytics/nn/modules/head.py,sha256=vlp3rMa54kjiuPqP32_RdgOb9KrHItiJx0ih1SFzQec,26853
-ultralytics/nn/modules/transformer.py,sha256=AxD9uURpCl-EqvXe3DiG6JW-pBzB16G-AahLdZ7yayo,17909
+ultralytics/nn/modules/transformer.py,sha256=8ux2-0ObrafMTYCLucLLVmqk9XWz74bwmWtJGDmgF6Q,18028
 ultralytics/nn/modules/utils.py,sha256=779QnnKp9v8jv251ESduTXJ0ol8HkIOLbGQWwEGQjhU,3196
 ultralytics/solutions/__init__.py,sha256=O_G9jh34NnFsHKSA8zcJH0CHtg1Q01JEiRWGwX3vGJY,631
 ultralytics/solutions/ai_gym.py,sha256=KQdx0RP9t9y1MqYMVlYUSn09SVJSUwKvgxPri_DhczM,4721
@@ -193,7 +204,7 @@ ultralytics/trackers/utils/__init__.py,sha256=mHtJuK4hwF8cuV-VHDc7tp6u6D1gHz2Z7J
 ultralytics/trackers/utils/gmc.py,sha256=-1oBNFRB-9EawJmUOT566AygLCVxJw-jsPSIOl5j_Hk,13683
 ultralytics/trackers/utils/kalman_filter.py,sha256=0oqhk59NKEiwcJ2FXnw6_sT4bIFC6Wu5IY2B-TGxJKU,15168
 ultralytics/trackers/utils/matching.py,sha256=UxhSGa5pN6WoYwYSBAkkt-O7xMxUR47VuUB6PfVNkb4,5404
-ultralytics/utils/__init__.py,sha256=FyRZRncXCAGmtFp1Dd4ACUK5zTrp6sOK-zFuG2Nwm1I,38905
+ultralytics/utils/__init__.py,sha256=kVU1d73Z6rSjgwyfdOwBaVrT2hYVwijUD54Wl_AWWA4,38942
 ultralytics/utils/autobatch.py,sha256=POJb9f8dioI7lPGnCc7bdxt0ncftXZa0bvOkip-XoWk,3969
 ultralytics/utils/benchmarks.py,sha256=6tdNcBLATllWpmAMUC6TW7DiCx1VKHhnQN4vkoqN3sE,23866
 ultralytics/utils/checks.py,sha256=hBkhOinWRzhpA5SbY1v-wCMdFeOemORRlmKBXgwoHYo,28498
@@ -206,9 +217,9 @@ ultralytics/utils/loss.py,sha256=mDHGmF-gjggAUVhI1dkCm7TtfZHCwz25XKm4M2xJKLs,339
 ultralytics/utils/metrics.py,sha256=UXMhBnTtMcpTANxmQqcYkVnj8NeAt39gZez0g6jbrW0,53786
 ultralytics/utils/ops.py,sha256=WJHyjyTH8xl5bRkBX0JB3K1sHAGONHx_joubUewE0A8,32709
 ultralytics/utils/patches.py,sha256=Oo3DkP7MbXnNGvPfoFSocAkVvaPh9kwMT_9RQUfjVhI,3594
-ultralytics/utils/plotting.py,sha256=5HRfiG2dklWZJheTxGTy0gFRk39utHcZbMJl7j2hnMI,55522
+ultralytics/utils/plotting.py,sha256=Bc-8SPs6R1BKMW1V8oVeD-ajMsWP0knAydsoFrB_doU,55522
 ultralytics/utils/tal.py,sha256=hia39MhWPFpDWOTAXC_5vz-9cUdiRHZs-UcTnxD4Dlo,16112
-ultralytics/utils/torch_utils.py,sha256=ClXfUpTKFkwD3Zfnjk9YnK0gsA-7YmvEasTDiVN9Wvw,28880
+ultralytics/utils/torch_utils.py,sha256=fvt3J2Oh1SgUcjUGSFK8sCKhCp826y6S7NBEiDGZpbI,28985
 ultralytics/utils/triton.py,sha256=gg1finxno_tY2Ge9PMhmu7PI9wvoFZoiicdT4Bhqv3w,3936
 ultralytics/utils/tuner.py,sha256=49KAadKZsUeCpwIm5Sn0grb0RPcMNI8vHGLwroDEJNI,6171
 ultralytics/utils/callbacks/__init__.py,sha256=YrWqC3BVVaTLob4iCPR6I36mUxIUOpPJW7B_LjT78Qw,214
@@ -222,9 +233,9 @@ ultralytics/utils/callbacks/neptune.py,sha256=5Z3ua5YBTUS56FH8VQKQG1aaIo9fH8GEyz
 ultralytics/utils/callbacks/raytune.py,sha256=ODVYzy-CoM4Uge0zjkh3Hnh9nF2M0vhDrSenXnvcizw,705
 ultralytics/utils/callbacks/tensorboard.py,sha256=QEgOVhUqY9akOs5TJIwz1Rvn6l32xWLpOxlwEyWF0B8,4136
 ultralytics/utils/callbacks/wb.py,sha256=9-fjQIdLjr3b73DTE3rHO171KvbH1VweJ-bmbv-rqTw,6747
-ultralytics-8.2.69.dist-info/LICENSE,sha256=DZak_2itbUtvHzD3E7GNUYSRK6jdOJ-GqncQ2weavLA,34523
-ultralytics-8.2.69.dist-info/METADATA,sha256=htZwlHV6f-WyWZpx2aAgEhKJYDRhK56EMOs0w0XwhZ4,41337
-ultralytics-8.2.69.dist-info/WHEEL,sha256=R0nc6qTxuoLk7ShA2_Y-UWkN8ZdfDBG2B6Eqpz2WXbs,91
-ultralytics-8.2.69.dist-info/entry_points.txt,sha256=YM_wiKyTe9yRrsEfqvYolNO5ngwfoL4-NwgKzc8_7sI,93
-ultralytics-8.2.69.dist-info/top_level.txt,sha256=XP49TwiMw4QGsvTLSYiJhz1xF_k7ev5mQ8jJXaXi45Q,12
-ultralytics-8.2.69.dist-info/RECORD,,
+ultralytics-8.2.71.dist-info/LICENSE,sha256=DZak_2itbUtvHzD3E7GNUYSRK6jdOJ-GqncQ2weavLA,34523
+ultralytics-8.2.71.dist-info/METADATA,sha256=neVOnXCAh1rp6O9ps9cdZbw_Pns6ZW0v8_Va4Prqy8k,41337
+ultralytics-8.2.71.dist-info/WHEEL,sha256=R0nc6qTxuoLk7ShA2_Y-UWkN8ZdfDBG2B6Eqpz2WXbs,91
+ultralytics-8.2.71.dist-info/entry_points.txt,sha256=YM_wiKyTe9yRrsEfqvYolNO5ngwfoL4-NwgKzc8_7sI,93
+ultralytics-8.2.71.dist-info/top_level.txt,sha256=XP49TwiMw4QGsvTLSYiJhz1xF_k7ev5mQ8jJXaXi45Q,12
+ultralytics-8.2.71.dist-info/RECORD,,