dgenerate-ultralytics-headless 8.3.196__py3-none-any.whl → 8.3.248__py3-none-any.whl

ultralytics/utils/autodevice.py

@@ -9,24 +9,23 @@ from ultralytics.utils.checks import check_requirements
 
 
 class GPUInfo:
-    """
-    Manages NVIDIA GPU information via pynvml with robust error handling.
+    """Manages NVIDIA GPU information via pynvml with robust error handling.
 
-    Provides methods to query detailed GPU statistics (utilization, memory, temp, power) and select the most idle
-    GPUs based on configurable criteria. It safely handles the absence or initialization failure of the pynvml
-    library by logging warnings and disabling related features, preventing application crashes.
+    Provides methods to query detailed GPU statistics (utilization, memory, temp, power) and select the most idle GPUs
+    based on configurable criteria. It safely handles the absence or initialization failure of the pynvml library by
+    logging warnings and disabling related features, preventing application crashes.
 
     Includes fallback logic using `torch.cuda` for basic device counting if NVML is unavailable during GPU
     selection. Manages NVML initialization and shutdown internally.
 
     Attributes:
         pynvml (module | None): The `pynvml` module if successfully imported and initialized, otherwise `None`.
-        nvml_available (bool): Indicates if `pynvml` is ready for use. True if import and `nvmlInit()` succeeded,
-            False otherwise.
-        gpu_stats (list[dict[str, Any]]): A list of dictionaries, each holding stats for one GPU. Populated on
-            initialization and by `refresh_stats()`. Keys include: 'index', 'name', 'utilization' (%),
-            'memory_used' (MiB), 'memory_total' (MiB), 'memory_free' (MiB), 'temperature' (C), 'power_draw' (W),
-            'power_limit' (W or 'N/A'). Empty if NVML is unavailable or queries fail.
+        nvml_available (bool): Indicates if `pynvml` is ready for use. True if import and `nvmlInit()` succeeded, False
+            otherwise.
+        gpu_stats (list[dict[str, Any]]): A list of dictionaries, each holding stats for one GPU, populated on
+        initialization and by `refresh_stats()`. Keys include: 'index', 'name', 'utilization' (%), 'memory_used' (MiB),
+            'memory_total' (MiB), 'memory_free' (MiB), 'temperature' (C), 'power_draw' (W), 'power_limit' (W or 'N/A').
+            Empty if NVML is unavailable or queries fail.
 
     Methods:
         refresh_stats: Refresh the internal gpu_stats list by querying NVML.
@@ -137,8 +136,7 @@ class GPUInfo:
     def select_idle_gpu(
         self, count: int = 1, min_memory_fraction: float = 0, min_util_fraction: float = 0
     ) -> list[int]:
-        """
-        Select the most idle GPUs based on utilization and free memory.
+        """Select the most idle GPUs based on utilization and free memory.
 
         Args:
             count (int): The number of idle GPUs to select.
@@ -154,9 +152,10 @@ class GPUInfo:
         """
         assert min_memory_fraction <= 1.0, f"min_memory_fraction must be <= 1.0, got {min_memory_fraction}"
         assert min_util_fraction <= 1.0, f"min_util_fraction must be <= 1.0, got {min_util_fraction}"
-        LOGGER.info(
-            f"Searching for {count} idle GPUs with free memory >= {min_memory_fraction * 100:.1f}% and free utilization >= {min_util_fraction * 100:.1f}%..."
+        criteria = (
+            f"free memory >= {min_memory_fraction * 100:.1f}% and free utilization >= {min_util_fraction * 100:.1f}%"
         )
+        LOGGER.info(f"Searching for {count} idle GPUs with {criteria}...")
 
         if count <= 0:
             return []
@@ -179,11 +178,11 @@ class GPUInfo:
         selected = [gpu["index"] for gpu in eligible_gpus[:count]]
 
         if selected:
+            if len(selected) < count:
+                LOGGER.warning(f"Requested {count} GPUs but only {len(selected)} met the idle criteria.")
             LOGGER.info(f"Selected idle CUDA devices {selected}")
         else:
-            LOGGER.warning(
-                f"No GPUs met criteria (Free Mem >= {min_memory_fraction * 100:.1f}% and Free Util >= {min_util_fraction * 100:.1f}%)."
-            )
+            LOGGER.warning(f"No GPUs met criteria ({criteria}).")
 
         return selected
 

ultralytics/utils/benchmarks.py

@@ -1,6 +1,6 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 """
-Benchmark a YOLO model formats for speed and accuracy.
+Benchmark YOLO model formats for speed and accuracy.
 
 Usage:
     from ultralytics.utils.benchmarks import ProfileModels, benchmark
@@ -25,6 +25,7 @@ MNN                     | `mnn`                     | yolo11n.mnn
 NCNN                    | `ncnn`                    | yolo11n_ncnn_model/
 IMX                     | `imx`                     | yolo11n_imx_model/
 RKNN                    | `rknn`                    | yolo11n_rknn_model/
+ExecuTorch              | `executorch`              | yolo11n_executorch_model/
 """
 
 from __future__ import annotations
@@ -43,7 +44,7 @@ import torch.cuda
 from ultralytics import YOLO, YOLOWorld
 from ultralytics.cfg import TASK2DATA, TASK2METRIC
 from ultralytics.engine.exporter import export_formats
-from ultralytics.utils import ARM64, ASSETS, IS_JETSON, LINUX, LOGGER, MACOS, TQDM, WEIGHTS_DIR, YAML
+from ultralytics.utils import ARM64, ASSETS, ASSETS_URL, IS_JETSON, LINUX, LOGGER, MACOS, TQDM, WEIGHTS_DIR, YAML
 from ultralytics.utils.checks import IS_PYTHON_3_13, check_imgsz, check_requirements, check_yolo, is_rockchip
 from ultralytics.utils.downloads import safe_download
 from ultralytics.utils.files import file_size
@@ -62,8 +63,7 @@ def benchmark(
     format="",
     **kwargs,
 ):
-    """
-    Benchmark a YOLO model across different formats for speed and accuracy.
+    """Benchmark a YOLO model across different formats for speed and accuracy.
 
     Args:
         model (str | Path): Path to the model file or directory.
@@ -78,8 +78,8 @@ def benchmark(
         **kwargs (Any): Additional keyword arguments for exporter.
 
     Returns:
-        (polars.DataFrame): A polars DataFrame with benchmark results for each format, including file size, metric,
-            and inference time.
+        (polars.DataFrame): A Polars DataFrame with benchmark results for each format, including file size, metric, and
+            inference time.
 
     Examples:
         Benchmark a YOLO model with default settings:
@@ -144,13 +144,18 @@ def benchmark(
             if format == "imx":
                 assert not is_end2end
                 assert not isinstance(model, YOLOWorld), "YOLOWorldv2 IMX exports not supported"
-                assert model.task == "detect", "IMX only supported for detection task"
-                assert "C2f" in model.__str__(), "IMX only supported for YOLOv8"  # TODO: enable for YOLO11
+                assert model.task in {"detect", "classify", "pose"}, (
+                    "IMX export is only supported for detection, classification and pose estimation tasks"
+                )
+                assert "C2f" in model.__str__(), "IMX only supported for YOLOv8n and YOLO11n"
             if format == "rknn":
                 assert not isinstance(model, YOLOWorld), "YOLOWorldv2 RKNN exports not supported yet"
                 assert not is_end2end, "End-to-end models not supported by RKNN yet"
                 assert LINUX, "RKNN only supported on Linux"
                 assert not is_rockchip(), "RKNN Inference only supported on Rockchip devices"
+            if format == "executorch":
+                assert not isinstance(model, YOLOWorld), "YOLOWorldv2 ExecuTorch exports not supported yet"
+                assert not is_end2end, "End-to-end models not supported by ExecuTorch yet"
             if "cpu" in device.type:
                 assert cpu, "inference not supported on CPU"
             if "cuda" in device.type:
@@ -170,6 +175,7 @@ def benchmark(
 
             # Predict
             assert model.task != "pose" or format != "pb", "GraphDef Pose inference is not supported"
+            assert model.task != "pose" or format != "executorch", "ExecuTorch Pose inference is not supported"
             assert format not in {"edgetpu", "tfjs"}, "inference not supported"
             assert format != "coreml" or platform.system() == "Darwin", "inference only supported on macOS>=10.13"
             if format == "ncnn":
@@ -220,8 +226,7 @@ def benchmark(
 
 
 class RF100Benchmark:
-    """
-    Benchmark YOLO model performance across various formats for speed and accuracy.
+    """Benchmark YOLO model performance across various formats for speed and accuracy.
 
     This class provides functionality to benchmark YOLO models on the RF100 dataset collection.
 
@@ -246,8 +251,7 @@ class RF100Benchmark:
         self.val_metrics = ["class", "images", "targets", "precision", "recall", "map50", "map95"]
 
     def set_key(self, api_key: str):
-        """
-        Set Roboflow API key for processing.
+        """Set Roboflow API key for processing.
 
         Args:
             api_key (str): The API key.
@@ -263,8 +267,7 @@ class RF100Benchmark:
         self.rf = Roboflow(api_key=api_key)
 
     def parse_dataset(self, ds_link_txt: str = "datasets_links.txt"):
-        """
-        Parse dataset links and download datasets.
+        """Parse dataset links and download datasets.
 
         Args:
             ds_link_txt (str): Path to the file containing dataset links.
@@ -281,12 +284,12 @@ class RF100Benchmark:
         (shutil.rmtree("rf-100"), os.mkdir("rf-100")) if os.path.exists("rf-100") else os.mkdir("rf-100")
         os.chdir("rf-100")
         os.mkdir("ultralytics-benchmarks")
-        safe_download("https://github.com/ultralytics/assets/releases/download/v0.0.0/datasets_links.txt")
+        safe_download(f"{ASSETS_URL}/datasets_links.txt")
 
         with open(ds_link_txt, encoding="utf-8") as file:
             for line in file:
                 try:
-                    _, url, workspace, project, version = re.split("/+", line.strip())
+                    _, _url, workspace, project, version = re.split("/+", line.strip())
                     self.ds_names.append(project)
                     proj_version = f"{project}-{version}"
                     if not Path(proj_version).exists():
@@ -308,8 +311,7 @@ class RF100Benchmark:
         YAML.dump(yaml_data, path)
 
     def evaluate(self, yaml_path: str, val_log_file: str, eval_log_file: str, list_ind: int):
-        """
-        Evaluate model performance on validation results.
+        """Evaluate model performance on validation results.
 
         Args:
             yaml_path (str): Path to the YAML configuration file.
@@ -357,7 +359,7 @@ class RF100Benchmark:
                     map_val = lst["map50"]
         else:
             LOGGER.info("Single dict found")
-            map_val = [res["map50"] for res in eval_lines][0]
+            map_val = next(res["map50"] for res in eval_lines)
 
         with open(eval_log_file, "a", encoding="utf-8") as f:
             f.write(f"{self.ds_names[list_ind]}: {map_val}\n")
@@ -366,8 +368,7 @@ class RF100Benchmark:
 
 
 class ProfileModels:
-    """
-    ProfileModels class for profiling different models on ONNX and TensorRT.
+    """ProfileModels class for profiling different models on ONNX and TensorRT.
 
     This class profiles the performance of different models, returning results such as model speed and FLOPs.
 
@@ -410,8 +411,7 @@ class ProfileModels:
         trt: bool = True,
         device: torch.device | str | None = None,
     ):
-        """
-        Initialize the ProfileModels class for profiling models.
+        """Initialize the ProfileModels class for profiling models.
 
         Args:
             paths (list[str]): List of paths of the models to be profiled.
@@ -425,12 +425,6 @@ class ProfileModels:
 
         Notes:
             FP16 'half' argument option removed for ONNX as slower on CPU than FP32.
-
-        Examples:
-            Initialize and profile models
-            >>> from ultralytics.utils.benchmarks import ProfileModels
-            >>> profiler = ProfileModels(["yolo11n.yaml", "yolov8s.yaml"], imgsz=640)
-            >>> profiler.run()
         """
         self.paths = paths
         self.num_timed_runs = num_timed_runs
@@ -442,8 +436,7 @@ class ProfileModels:
         self.device = device if isinstance(device, torch.device) else select_device(device)
 
     def run(self):
-        """
-        Profile YOLO models for speed and accuracy across various formats including ONNX and TensorRT.
+        """Profile YOLO models for speed and accuracy across various formats including ONNX and TensorRT.
 
         Returns:
             (list[dict]): List of dictionaries containing profiling results for each model.
@@ -497,8 +490,7 @@ class ProfileModels:
         return output
 
     def get_files(self):
-        """
-        Return a list of paths for all relevant model files given by the user.
+        """Return a list of paths for all relevant model files given by the user.
 
         Returns:
             (list[Path]): List of Path objects for the model files.
@@ -524,8 +516,7 @@ class ProfileModels:
 
     @staticmethod
     def iterative_sigma_clipping(data: np.ndarray, sigma: float = 2, max_iters: int = 3):
-        """
-        Apply iterative sigma clipping to data to remove outliers.
+        """Apply iterative sigma clipping to data to remove outliers.
 
         Args:
             data (np.ndarray): Input data array.
@@ -545,8 +536,7 @@ class ProfileModels:
         return data
 
     def profile_tensorrt_model(self, engine_file: str, eps: float = 1e-3):
-        """
-        Profile YOLO model performance with TensorRT, measuring average run time and standard deviation.
+        """Profile YOLO model performance with TensorRT, measuring average run time and standard deviation.
 
         Args:
             engine_file (str): Path to the TensorRT engine file.
@@ -583,9 +573,13 @@ class ProfileModels:
         run_times = self.iterative_sigma_clipping(np.array(run_times), sigma=2, max_iters=3)  # sigma clipping
         return np.mean(run_times), np.std(run_times)
 
+    @staticmethod
+    def check_dynamic(tensor_shape):
+        """Check whether the tensor shape in the ONNX model is dynamic."""
+        return not all(isinstance(dim, int) and dim >= 0 for dim in tensor_shape)
+
     def profile_onnx_model(self, onnx_file: str, eps: float = 1e-3):
-        """
-        Profile an ONNX model, measuring average inference time and standard deviation across multiple runs.
+        """Profile an ONNX model, measuring average inference time and standard deviation across multiple runs.
 
         Args:
             onnx_file (str): Path to the ONNX model file.
@@ -595,7 +589,7 @@ class ProfileModels:
             mean_time (float): Mean inference time in milliseconds.
             std_time (float): Standard deviation of inference time in milliseconds.
         """
-        check_requirements("onnxruntime")
+        check_requirements([("onnxruntime", "onnxruntime-gpu")])  # either package meets requirements
         import onnxruntime as ort
 
         # Session with either 'TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider'
@@ -604,27 +598,36 @@ class ProfileModels:
         sess_options.intra_op_num_threads = 8  # Limit the number of threads
         sess = ort.InferenceSession(onnx_file, sess_options, providers=["CPUExecutionProvider"])
 
-        input_tensor = sess.get_inputs()[0]
-        input_type = input_tensor.type
-        dynamic = not all(isinstance(dim, int) and dim >= 0 for dim in input_tensor.shape)  # dynamic input shape
-        input_shape = (1, 3, self.imgsz, self.imgsz) if dynamic else input_tensor.shape
-
-        # Mapping ONNX datatype to numpy datatype
-        if "float16" in input_type:
-            input_dtype = np.float16
-        elif "float" in input_type:
-            input_dtype = np.float32
-        elif "double" in input_type:
-            input_dtype = np.float64
-        elif "int64" in input_type:
-            input_dtype = np.int64
-        elif "int32" in input_type:
-            input_dtype = np.int32
-        else:
-            raise ValueError(f"Unsupported ONNX datatype {input_type}")
+        input_data_dict = {}
+        for input_tensor in sess.get_inputs():
+            input_type = input_tensor.type
+            if self.check_dynamic(input_tensor.shape):
+                if len(input_tensor.shape) != 4 and self.check_dynamic(input_tensor.shape[1:]):
+                    raise ValueError(f"Unsupported dynamic shape {input_tensor.shape} of {input_tensor.name}")
+                input_shape = (
+                    (1, 3, self.imgsz, self.imgsz) if len(input_tensor.shape) == 4 else (1, *input_tensor.shape[1:])
+                )
+            else:
+                input_shape = input_tensor.shape
+
+            # Mapping ONNX datatype to numpy datatype
+            if "float16" in input_type:
+                input_dtype = np.float16
+            elif "float" in input_type:
+                input_dtype = np.float32
+            elif "double" in input_type:
+                input_dtype = np.float64
+            elif "int64" in input_type:
+                input_dtype = np.int64
+            elif "int32" in input_type:
+                input_dtype = np.int32
+            else:
+                raise ValueError(f"Unsupported ONNX datatype {input_type}")
+
+            input_data = np.random.rand(*input_shape).astype(input_dtype)
+            input_name = input_tensor.name
+            input_data_dict[input_name] = input_data
 
-        input_data = np.random.rand(*input_shape).astype(input_dtype)
-        input_name = input_tensor.name
         output_name = sess.get_outputs()[0].name
 
         # Warmup runs
@@ -632,7 +635,7 @@ class ProfileModels:
         for _ in range(3):
             start_time = time.time()
             for _ in range(self.num_warmup_runs):
-                sess.run([output_name], {input_name: input_data})
+                sess.run([output_name], input_data_dict)
             elapsed = time.time() - start_time
 
         # Compute number of runs as higher of min_time or num_timed_runs
@@ -642,7 +645,7 @@ class ProfileModels:
         run_times = []
         for _ in TQDM(range(num_runs), desc=onnx_file):
             start_time = time.time()
-            sess.run([output_name], {input_name: input_data})
+            sess.run([output_name], input_data_dict)
             run_times.append((time.time() - start_time) * 1000)  # Convert to milliseconds
 
         run_times = self.iterative_sigma_clipping(np.array(run_times), sigma=2, max_iters=5)  # sigma clipping
@@ -655,8 +658,7 @@ class ProfileModels:
         t_engine: tuple[float, float],
         model_info: tuple[float, float, float, float],
     ):
-        """
-        Generate a table row string with model performance metrics.
+        """Generate a table row string with model performance metrics.
 
         Args:
             model_name (str): Name of the model.
@@ -667,7 +669,7 @@ class ProfileModels:
         Returns:
             (str): Formatted table row string with model metrics.
         """
-        layers, params, gradients, flops = model_info
+        _layers, params, _gradients, flops = model_info
         return (
             f"| {model_name:18s} | {self.imgsz} | - | {t_onnx[0]:.1f}±{t_onnx[1]:.1f} ms | {t_engine[0]:.1f}±"
             f"{t_engine[1]:.1f} ms | {params / 1e6:.1f} | {flops:.1f} |"
@@ -680,8 +682,7 @@ class ProfileModels:
         t_engine: tuple[float, float],
         model_info: tuple[float, float, float, float],
     ):
-        """
-        Generate a dictionary of profiling results.
+        """Generate a dictionary of profiling results.
 
         Args:
             model_name (str): Name of the model.
@@ -692,7 +693,7 @@ class ProfileModels:
         Returns:
             (dict): Dictionary containing profiling results.
         """
-        layers, params, gradients, flops = model_info
+        _layers, params, _gradients, flops = model_info
         return {
             "model/name": model_name,
             "model/parameters": params,
@@ -703,8 +704,7 @@ class ProfileModels:
 
     @staticmethod
     def print_table(table_rows: list[str]):
-        """
-        Print a formatted table of model profiling results.
+        """Print a formatted table of model profiling results.
 
         Args:
             table_rows (list[str]): List of formatted table row strings.

ultralytics/utils/callbacks/base.py

@@ -175,13 +175,12 @@ default_callbacks = {
 
 
 def get_default_callbacks():
-    """
-    Get the default callbacks for Ultralytics training, validation, prediction, and export processes.
+    """Get the default callbacks for Ultralytics training, validation, prediction, and export processes.
 
     Returns:
         (dict): Dictionary of default callbacks for various training events. Each key represents an event during the
-            training process, and the corresponding value is a list of callback functions executed when that event
-            occurs.
+            training process, and the corresponding value is a list of callback functions executed when that
+            event occurs.
 
     Examples:
         >>> callbacks = get_default_callbacks()
@@ -192,17 +191,16 @@ def get_default_callbacks():
 
 
 def add_integration_callbacks(instance):
-    """
-    Add integration callbacks to the instance's callbacks dictionary.
+    """Add integration callbacks to the instance's callbacks dictionary.
 
     This function loads and adds various integration callbacks to the provided instance. The specific callbacks added
     depend on the type of instance provided. All instances receive HUB callbacks, while Trainer instances also receive
-    additional callbacks for various integrations like ClearML, Comet, DVC, MLflow, Neptune, Ray Tune, TensorBoard,
-    and Weights & Biases.
+    additional callbacks for various integrations like ClearML, Comet, DVC, MLflow, Neptune, Ray Tune, TensorBoard, and
+    Weights & Biases.
 
     Args:
-        instance (Trainer | Predictor | Validator | Exporter): The object instance to which callbacks will be added.
-            The type of instance determines which callbacks are loaded.
+        instance (Trainer | Predictor | Validator | Exporter): The object instance to which callbacks will be added. The
+            type of instance determines which callbacks are loaded.
 
     Examples:
         >>> from ultralytics.engine.trainer import BaseTrainer

ultralytics/utils/callbacks/clearml.py

@@ -15,8 +15,7 @@ except (ImportError, AssertionError):
 
 
 def _log_debug_samples(files, title: str = "Debug Samples") -> None:
-    """
-    Log files (images) as debug samples in the ClearML task.
+    """Log files (images) as debug samples in the ClearML task.
 
     Args:
         files (list[Path]): A list of file paths in PosixPath format.
@@ -35,8 +34,7 @@ def _log_debug_samples(files, title: str = "Debug Samples") -> None:
 
 
 def _log_plot(title: str, plot_path: str) -> None:
-    """
-    Log an image as a plot in the plot section of ClearML.
+    """Log an image as a plot in the plot section of ClearML.
 
     Args:
         title (str): The title of the plot.
@@ -125,15 +123,9 @@ def on_train_end(trainer) -> None:
     """Log final model and training results on training completion."""
     if task := Task.current_task():
         # Log final results, confusion matrix and PR plots
-        files = [
-            "results.png",
-            "confusion_matrix.png",
-            "confusion_matrix_normalized.png",
-            *(f"{x}_curve.png" for x in ("F1", "PR", "P", "R")),
-        ]
-        files = [(trainer.save_dir / f) for f in files if (trainer.save_dir / f).exists()]  # filter existing files
-        for f in files:
-            _log_plot(title=f.stem, plot_path=f)
+        for f in [*trainer.plots.keys(), *trainer.validator.plots.keys()]:
+            if "batch" not in f.name:
+                _log_plot(title=f.stem, plot_path=f)
         # Report final metrics
         for k, v in trainer.validator.metrics.results_dict.items():
             task.get_logger().report_single_value(k, v)