ultralytics 8.3.105__tar.gz → 8.3.107__tar.gz

{ultralytics-8.3.105/ultralytics.egg-info → ultralytics-8.3.107}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ultralytics
-Version: 8.3.105
+Version: 8.3.107
 Summary: Ultralytics YOLO 🚀 for SOTA object detection, multi-object tracking, instance segmentation, pose estimation and image classification.
 Author-email: Glenn Jocher <glenn.jocher@ultralytics.com>, Jing Qiu <jing.qiu@ultralytics.com>
 Maintainer-email: Ultralytics <hello@ultralytics.com>
@@ -62,7 +62,7 @@ Provides-Extra: export
 Requires-Dist: onnx>=1.12.0; extra == "export"
 Requires-Dist: coremltools>=8.0; (platform_system != "Windows" and python_version <= "3.12") and extra == "export"
 Requires-Dist: scikit-learn>=1.3.2; (platform_system != "Windows" and python_version <= "3.12") and extra == "export"
-Requires-Dist: openvino!=2025.0.0,>=2024.0.0; extra == "export"
+Requires-Dist: openvino>=2024.0.0; extra == "export"
 Requires-Dist: tensorflow>=2.0.0; extra == "export"
 Requires-Dist: tensorflowjs>=4.0.0; extra == "export"
 Requires-Dist: tensorstore>=0.1.63; (platform_machine == "aarch64" and python_version >= "3.9") and extra == "export"

{ultralytics-8.3.105 → ultralytics-8.3.107}/pyproject.toml

@@ -96,7 +96,7 @@ export = [
     "onnx>=1.12.0", # ONNX export
     "coremltools>=8.0; platform_system != 'Windows' and python_version <= '3.12'", # CoreML supported on macOS and Linux
     "scikit-learn>=1.3.2; platform_system != 'Windows' and python_version <= '3.12'", # CoreML k-means quantization
-    "openvino>=2024.0.0,!=2025.0.0", # OpenVINO export
+    "openvino>=2024.0.0",  # OpenVINO export
     "tensorflow>=2.0.0", # TF bug https://github.com/ultralytics/ultralytics/issues/5161
     "tensorflowjs>=4.0.0", # TF.js export, automatically installs tensorflow
     "tensorstore>=0.1.63; platform_machine == 'aarch64' and python_version >= '3.9'", # for TF Raspberry Pi exports
@@ -134,7 +134,7 @@ ultralytics = "ultralytics.cfg:entrypoint"
 # Tools settings -------------------------------------------------------------------------------------------------------
 [tool.setuptools]  # configuration specific to the `setuptools` build backend.
 packages = { find = { where = ["."], include = ["ultralytics", "ultralytics.*"] } }
-package-data = { "ultralytics" = ["**/*.yaml", "../tests/*.py"], "ultralytics.assets" = ["*.jpg"] }
+package-data = { "ultralytics" = ["**/*.yaml"], "ultralytics.assets" = ["*.jpg"] }
 
 [tool.setuptools.dynamic]
 version = { attr = "ultralytics.__version__" }

{ultralytics-8.3.105 → ultralytics-8.3.107}/tests/test_solutions.py

@@ -153,11 +153,11 @@ def process_video(solution, video_path, needs_frame_count=False):
     cap.release()
 
 
-@pytest.mark.slow
 @pytest.mark.parametrize("name, solution_class, needs_frame_count, video, kwargs", SOLUTIONS)
 def test_solution(name, solution_class, needs_frame_count, video, kwargs):
     """Test individual Ultralytics solution."""
-    safe_download(url=f"{ASSETS_URL}/{video}", dir=TMP)
+    if video:
+        safe_download(url=f"{ASSETS_URL}/{video}", dir=TMP)
     if name == "ParkingManager":
         safe_download(url=f"{ASSETS_URL}/{PARKING_AREAS_JSON}", dir=TMP)
         safe_download(url=f"{ASSETS_URL}/{PARKING_MODEL}", dir=TMP)

{ultralytics-8.3.105 → ultralytics-8.3.107}/ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 
-__version__ = "8.3.105"
+__version__ = "8.3.107"
 
 import os
 

{ultralytics-8.3.105 → ultralytics-8.3.107}/ultralytics/engine/exporter.py

@@ -55,7 +55,6 @@ TensorFlow.js:
     $ npm start
 """
 
-import gc
 import json
 import os
 import re
@@ -86,7 +85,7 @@ from ultralytics.utils import (
     LINUX,
     LOGGER,
     MACOS,
-    PYTHON_VERSION,
+    MACOS_VERSION,
     RKNN_CHIPS,
     ROOT,
     WINDOWS,
@@ -104,6 +103,7 @@ from ultralytics.utils.checks import (
     is_sudo_available,
 )
 from ultralytics.utils.downloads import attempt_download_asset, get_github_assets, safe_download
+from ultralytics.utils.export import export_engine, export_onnx
 from ultralytics.utils.files import file_size, spaces_in_path
 from ultralytics.utils.ops import Profile, nms_rotated, xywh2xyxy
 from ultralytics.utils.torch_utils import TORCH_1_13, get_latest_opset, select_device
@@ -578,16 +578,14 @@ class Exporter:
             check_requirements("onnxslim>=0.1.46")  # Older versions has bug with OBB
 
         with arange_patch(self.args):
-            torch.onnx.export(
+            export_onnx(
                 NMSModel(self.model, self.args) if self.args.nms else self.model,
                 self.im,
                 f,
-                verbose=False,
-                opset_version=opset_version,
-                do_constant_folding=True,  # WARNING: DNN inference with torch>=1.12 may require do_constant_folding=False
+                opset=opset_version,
                 input_names=["images"],
                 output_names=output_names,
-                dynamic_axes=dynamic or None,
+                dynamic=dynamic or None,
             )
 
         # Checks
@@ -615,7 +613,10 @@ class Exporter:
     @try_export
     def export_openvino(self, prefix=colorstr("OpenVINO:")):
         """YOLO OpenVINO export."""
-        check_requirements("openvino>=2024.0.0,!=2025.0.0")
+        if MACOS:
+            msg = "OpenVINO error in macOS>=15.4 https://github.com/openvinotoolkit/openvino/issues/30023"
+            check_version(MACOS_VERSION, "<15.4", name="macOS ", hard=True, msg=msg)
+        check_requirements("openvino>=2024.0.0")
         import openvino as ov
 
         LOGGER.info(f"\n{prefix} starting export with openvino {ov.__version__}...")
@@ -795,7 +796,7 @@ class Exporter:
     def export_coreml(self, prefix=colorstr("CoreML:")):
         """YOLO CoreML export."""
         mlmodel = self.args.format.lower() == "mlmodel"  # legacy *.mlmodel export format requested
-        check_requirements("coremltools>=6.0,<=6.2" if mlmodel else "coremltools>=8.0")
+        check_requirements("coremltools>=8.0")
         import coremltools as ct  # noqa
 
         LOGGER.info(f"\n{prefix} starting export with coremltools {ct.__version__}...")
@@ -819,11 +820,15 @@ class Exporter:
                 # TODO CoreML Segment and Pose model pipelining
             model = self.model
         ts = torch.jit.trace(model.eval(), self.im, strict=False)  # TorchScript model
+
+        # Based on apple's documentation it is better to leave out the minimum_deployment target and let that get set
+        # Internally based on the model conversion and output type.
+        # Setting minimum_depoloyment_target >= iOS16 will require setting compute_precision=ct.precision.FLOAT32.
+        # iOS16 adds in better support for FP16, but none of the CoreML NMS specifications handle FP16 as input.
         ct_model = ct.convert(
             ts,
             inputs=[ct.ImageType("image", shape=self.im.shape, scale=scale, bias=bias)],  # expects ct.TensorType
             classifier_config=classifier_config,
-            minimum_deployment_target=ct.target.iOS15,  # warning: >=16 causes pipeline errors
             convert_to="neuralnetwork" if mlmodel else "mlprogram",
         )
         bits, mode = (8, "kmeans") if self.args.int8 else (16, "linear") if self.args.half else (32, None)
@@ -840,8 +845,6 @@ class Exporter:
                 ct_model = cto.palettize_weights(ct_model, config=config)
         if self.args.nms and self.model.task == "detect":
             if mlmodel:
-                # coremltools<=6.2 NMS export requires Python<3.11
-                check_version(PYTHON_VERSION, "<3.11", name="Python ", hard=True)
                 weights_dir = None
             else:
                 ct_model.save(str(f))  # save otherwise weights_dir does not exist
@@ -882,134 +885,22 @@ class Exporter:
 
         # Setup and checks
         LOGGER.info(f"\n{prefix} starting export with TensorRT {trt.__version__}...")
-        is_trt10 = int(trt.__version__.split(".")[0]) >= 10  # is TensorRT >= 10
         assert Path(f_onnx).exists(), f"failed to export ONNX file: {f_onnx}"
         f = self.file.with_suffix(".engine")  # TensorRT engine file
-        logger = trt.Logger(trt.Logger.INFO)
-        if self.args.verbose:
-            logger.min_severity = trt.Logger.Severity.VERBOSE
-
-        # Engine builder
-        builder = trt.Builder(logger)
-        config = builder.create_builder_config()
-        workspace = int((self.args.workspace or 0) * (1 << 30))
-        if is_trt10 and workspace > 0:
-            config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, workspace)
-        elif workspace > 0:  # TensorRT versions 7, 8
-            config.max_workspace_size = workspace
-        flag = 1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH)
-        network = builder.create_network(flag)
-        half = builder.platform_has_fast_fp16 and self.args.half
-        int8 = builder.platform_has_fast_int8 and self.args.int8
-
-        # Optionally switch to DLA if enabled
-        if dla is not None:
-            if not IS_JETSON:
-                raise ValueError("DLA is only available on NVIDIA Jetson devices")
-            LOGGER.info(f"{prefix} enabling DLA on core {dla}...")
-            if not self.args.half and not self.args.int8:
-                raise ValueError(
-                    "DLA requires either 'half=True' (FP16) or 'int8=True' (INT8) to be enabled. Please enable one of them and try again."
-                )
-            config.default_device_type = trt.DeviceType.DLA
-            config.DLA_core = int(dla)
-            config.set_flag(trt.BuilderFlag.GPU_FALLBACK)
-
-        # Read ONNX file
-        parser = trt.OnnxParser(network, logger)
-        if not parser.parse_from_file(f_onnx):
-            raise RuntimeError(f"failed to load ONNX file: {f_onnx}")
-
-        # Network inputs
-        inputs = [network.get_input(i) for i in range(network.num_inputs)]
-        outputs = [network.get_output(i) for i in range(network.num_outputs)]
-        for inp in inputs:
-            LOGGER.info(f'{prefix} input "{inp.name}" with shape{inp.shape} {inp.dtype}')
-        for out in outputs:
-            LOGGER.info(f'{prefix} output "{out.name}" with shape{out.shape} {out.dtype}')
-
-        if self.args.dynamic:
-            shape = self.im.shape
-            if shape[0] <= 1:
-                LOGGER.warning(f"{prefix} WARNING ⚠️ 'dynamic=True' model requires max batch size, i.e. 'batch=16'")
-            profile = builder.create_optimization_profile()
-            min_shape = (1, shape[1], 32, 32)  # minimum input shape
-            max_shape = (*shape[:2], *(int(max(1, self.args.workspace or 1) * d) for d in shape[2:]))  # max input shape
-            for inp in inputs:
-                profile.set_shape(inp.name, min=min_shape, opt=shape, max=max_shape)
-            config.add_optimization_profile(profile)
-
-        LOGGER.info(f"{prefix} building {'INT8' if int8 else 'FP' + ('16' if half else '32')} engine as {f}")
-        if int8:
-            config.set_flag(trt.BuilderFlag.INT8)
-            config.set_calibration_profile(profile)
-            config.profiling_verbosity = trt.ProfilingVerbosity.DETAILED
-
-            class EngineCalibrator(trt.IInt8Calibrator):
-                def __init__(
-                    self,
-                    dataset,  # ultralytics.data.build.InfiniteDataLoader
-                    batch: int,
-                    cache: str = "",
-                ) -> None:
-                    trt.IInt8Calibrator.__init__(self)
-                    self.dataset = dataset
-                    self.data_iter = iter(dataset)
-                    self.algo = trt.CalibrationAlgoType.ENTROPY_CALIBRATION_2
-                    self.batch = batch
-                    self.cache = Path(cache)
-
-                def get_algorithm(self) -> trt.CalibrationAlgoType:
-                    """Get the calibration algorithm to use."""
-                    return self.algo
-
-                def get_batch_size(self) -> int:
-                    """Get the batch size to use for calibration."""
-                    return self.batch or 1
-
-                def get_batch(self, names) -> list:
-                    """Get the next batch to use for calibration, as a list of device memory pointers."""
-                    try:
-                        im0s = next(self.data_iter)["img"] / 255.0
-                        im0s = im0s.to("cuda") if im0s.device.type == "cpu" else im0s
-                        return [int(im0s.data_ptr())]
-                    except StopIteration:
-                        # Return [] or None, signal to TensorRT there is no calibration data remaining
-                        return None
-
-                def read_calibration_cache(self) -> bytes:
-                    """Use existing cache instead of calibrating again, otherwise, implicitly return None."""
-                    if self.cache.exists() and self.cache.suffix == ".cache":
-                        return self.cache.read_bytes()
-
-                def write_calibration_cache(self, cache) -> None:
-                    """Write calibration cache to disk."""
-                    _ = self.cache.write_bytes(cache)
-
-            # Load dataset w/ builder (for batching) and calibrate
-            config.int8_calibrator = EngineCalibrator(
-                dataset=self.get_int8_calibration_dataloader(prefix),
-                batch=2 * self.args.batch,  # TensorRT INT8 calibration should use 2x batch size
-                cache=str(self.file.with_suffix(".cache")),
-            )
-
-        elif half:
-            config.set_flag(trt.BuilderFlag.FP16)
-
-        # Free CUDA memory
-        del self.model
-        gc.collect()
-        torch.cuda.empty_cache()
-
-        # Write file
-        build = builder.build_serialized_network if is_trt10 else builder.build_engine
-        with build(network, config) as engine, open(f, "wb") as t:
-            # Metadata
-            meta = json.dumps(self.metadata)
-            t.write(len(meta).to_bytes(4, byteorder="little", signed=True))
-            t.write(meta.encode())
-            # Model
-            t.write(engine if is_trt10 else engine.serialize())
+        export_engine(
+            f_onnx,
+            f,
+            self.args.workspace,
+            self.args.half,
+            self.args.int8,
+            self.args.dynamic,
+            self.im.shape,
+            dla=dla,
+            dataset=self.get_int8_calibration_dataloader(prefix) if self.args.int8 else None,
+            metadata=self.metadata,
+            verbose=self.args.verbose,
+            prefix=prefix,
+        )
 
         return f, None
 
@@ -1469,7 +1360,7 @@ class Exporter:
 
         # 3. Create NMS protobuf
         nms_spec = ct.proto.Model_pb2.Model()
-        nms_spec.specificationVersion = 9
+        nms_spec.specificationVersion = spec.specificationVersion
         for i in range(2):
             decoder_output = model._spec.description.output[i].SerializeToString()
             nms_spec.description.input.add()
@@ -1522,7 +1413,7 @@ class Exporter:
         pipeline.spec.description.output[1].ParseFromString(nms_model._spec.description.output[1].SerializeToString())
 
         # Update metadata
-        pipeline.spec.specificationVersion = 9
+        pipeline.spec.specificationVersion = spec.specificationVersion
         pipeline.spec.description.metadata.userDefined.update(
             {"IoU threshold": str(nms.iouThreshold), "Confidence threshold": str(nms.confidenceThreshold)}
         )

{ultralytics-8.3.105 → ultralytics-8.3.107}/ultralytics/nn/autobackend.py

@@ -258,7 +258,7 @@ class AutoBackend(nn.Module):
         # OpenVINO
         elif xml:
             LOGGER.info(f"Loading {w} for OpenVINO inference...")
-            check_requirements("openvino>=2024.0.0,!=2025.0.0")
+            check_requirements("openvino>=2024.0.0")
             import openvino as ov
 
             core = ov.Core()
@@ -511,9 +511,9 @@ class AutoBackend(nn.Module):
             if not w.is_file():  # if not *.rknn
                 w = next(w.rglob("*.rknn"))  # get *.rknn file from *_rknn_model dir
             rknn_model = RKNNLite()
-            rknn_model.load_rknn(w)
+            rknn_model.load_rknn(str(w))
             rknn_model.init_runtime()
-            metadata = Path(w).parent / "metadata.yaml"
+            metadata = w.parent / "metadata.yaml"
 
         # Any other format (unsupported)
         else:

{ultralytics-8.3.105 → ultralytics-8.3.107}/ultralytics/utils/__init__.py

@@ -48,6 +48,7 @@ VERBOSE = str(os.getenv("YOLO_VERBOSE", True)).lower() == "true"  # global verbo
 TQDM_BAR_FORMAT = "{l_bar}{bar:10}{r_bar}" if VERBOSE else None  # tqdm bar format
 LOGGING_NAME = "ultralytics"
 MACOS, LINUX, WINDOWS = (platform.system() == x for x in ["Darwin", "Linux", "Windows"])  # environment booleans
+MACOS_VERSION = platform.mac_ver()[0] if MACOS else None
 ARM64 = platform.machine() in {"arm64", "aarch64"}  # ARM64 booleans
 PYTHON_VERSION = platform.python_version()
 TORCH_VERSION = torch.__version__

ultralytics-8.3.107/ultralytics/utils/export.py

@@ -0,0 +1,217 @@
+import json
+from pathlib import Path
+
+import torch
+
+from ultralytics.utils import IS_JETSON, LOGGER
+
+
+def export_onnx(
+    torch_model,
+    im,
+    onnx_file,
+    opset=14,
+    input_names=["images"],
+    output_names=["output0"],
+    dynamic=False,
+):
+    """
+    Exports a PyTorch model to ONNX format.
+
+    Args:
+        torch_model (torch.nn.Module): The PyTorch model to export.
+        im (torch.Tensor): Example input tensor for the model.
+        onnx_file (str): Path to save the exported ONNX file.
+        opset (int): ONNX opset version to use for export.
+        input_names (list): List of input tensor names.
+        output_names (list): List of output tensor names.
+        dynamic (bool | dict, optional): Whether to enable dynamic axes. Defaults to False.
+
+    Notes:
+        - Setting `do_constant_folding=True` may cause issues with DNN inference for torch>=1.12.
+    """
+    torch.onnx.export(
+        torch_model,
+        im,
+        onnx_file,
+        verbose=False,
+        opset_version=opset,
+        do_constant_folding=True,  # WARNING: DNN inference with torch>=1.12 may require do_constant_folding=False
+        input_names=input_names,
+        output_names=output_names,
+        dynamic_axes=dynamic or None,
+    )
+
+
+def export_engine(
+    onnx_file,
+    engine_file=None,
+    workspace=None,
+    half=False,
+    int8=False,
+    dynamic=False,
+    shape=(1, 3, 640, 640),
+    dla=None,
+    dataset=None,
+    metadata=None,
+    verbose=False,
+    prefix="",
+):
+    """
+    Exports a YOLO model to TensorRT engine format.
+
+    Args:
+        onnx_file (str): Path to the ONNX file to be converted.
+        engine_file (str, optional): Path to save the generated TensorRT engine file.
+        workspace (int, optional): Workspace size in GB for TensorRT. Defaults to None.
+        half (bool, optional): Enable FP16 precision. Defaults to False.
+        int8 (bool, optional): Enable INT8 precision. Defaults to False.
+        dynamic (bool, optional): Enable dynamic input shapes. Defaults to False.
+        shape (tuple, optional): Input shape (batch, channels, height, width). Defaults to (1, 3, 640, 640).
+        dla (int, optional): DLA core to use (Jetson devices only). Defaults to None.
+        dataset (ultralytics.data.build.InfiniteDataLoader, optional): Dataset for INT8 calibration. Defaults to None.
+        metadata (dict, optional): Metadata to include in the engine file. Defaults to None.
+        verbose (bool, optional): Enable verbose logging. Defaults to False.
+        prefix (str, optional): Prefix for log messages. Defaults to "".
+
+    Raises:
+        ValueError: If DLA is enabled on non-Jetson devices or required precision is not set.
+        RuntimeError: If the ONNX file cannot be parsed.
+
+    Notes:
+        - TensorRT version compatibility is handled for workspace size and engine building.
+        - INT8 calibration requires a dataset and generates a calibration cache.
+        - Metadata is serialized and written to the engine file if provided.
+    """
+    import tensorrt as trt  # noqa
+
+    engine_file = engine_file or Path(onnx_file).with_suffix(".engine")
+
+    logger = trt.Logger(trt.Logger.INFO)
+    if verbose:
+        logger.min_severity = trt.Logger.Severity.VERBOSE
+
+    # Engine builder
+    builder = trt.Builder(logger)
+    config = builder.create_builder_config()
+    workspace = int((workspace or 0) * (1 << 30))
+    is_trt10 = int(trt.__version__.split(".")[0]) >= 10  # is TensorRT >= 10
+    if is_trt10 and workspace > 0:
+        config.set_memory_pool_limit(trt.MemoryPoolType.WORKSPACE, workspace)
+    elif workspace > 0:  # TensorRT versions 7, 8
+        config.max_workspace_size = workspace
+    flag = 1 << int(trt.NetworkDefinitionCreationFlag.EXPLICIT_BATCH)
+    network = builder.create_network(flag)
+    half = builder.platform_has_fast_fp16 and half
+    int8 = builder.platform_has_fast_int8 and int8
+
+    # Optionally switch to DLA if enabled
+    if dla is not None:
+        if not IS_JETSON:
+            raise ValueError("DLA is only available on NVIDIA Jetson devices")
+        LOGGER.info(f"{prefix} enabling DLA on core {dla}...")
+        if not half and not int8:
+            raise ValueError(
+                "DLA requires either 'half=True' (FP16) or 'int8=True' (INT8) to be enabled. Please enable one of them and try again."
+            )
+        config.default_device_type = trt.DeviceType.DLA
+        config.DLA_core = int(dla)
+        config.set_flag(trt.BuilderFlag.GPU_FALLBACK)
+
+    # Read ONNX file
+    parser = trt.OnnxParser(network, logger)
+    if not parser.parse_from_file(onnx_file):
+        raise RuntimeError(f"failed to load ONNX file: {onnx_file}")
+
+    # Network inputs
+    inputs = [network.get_input(i) for i in range(network.num_inputs)]
+    outputs = [network.get_output(i) for i in range(network.num_outputs)]
+    for inp in inputs:
+        LOGGER.info(f'{prefix} input "{inp.name}" with shape{inp.shape} {inp.dtype}')
+    for out in outputs:
+        LOGGER.info(f'{prefix} output "{out.name}" with shape{out.shape} {out.dtype}')
+
+    if dynamic:
+        if shape[0] <= 1:
+            LOGGER.warning(f"{prefix} WARNING ⚠️ 'dynamic=True' model requires max batch size, i.e. 'batch=16'")
+        profile = builder.create_optimization_profile()
+        min_shape = (1, shape[1], 32, 32)  # minimum input shape
+        max_shape = (*shape[:2], *(int(max(1, workspace or 1) * d) for d in shape[2:]))  # max input shape
+        for inp in inputs:
+            profile.set_shape(inp.name, min=min_shape, opt=shape, max=max_shape)
+        config.add_optimization_profile(profile)
+
+    LOGGER.info(f"{prefix} building {'INT8' if int8 else 'FP' + ('16' if half else '32')} engine as {engine_file}")
+    if int8:
+        config.set_flag(trt.BuilderFlag.INT8)
+        config.set_calibration_profile(profile)
+        config.profiling_verbosity = trt.ProfilingVerbosity.DETAILED
+
+        class EngineCalibrator(trt.IInt8Calibrator):
+            """
+            Custom INT8 calibrator for TensorRT.
+
+            Args:
+                dataset (object): Dataset for calibration.
+                batch (int): Batch size for calibration.
+                cache (str, optional): Path to save the calibration cache. Defaults to "".
+            """
+
+            def __init__(
+                self,
+                dataset,  # ultralytics.data.build.InfiniteDataLoader
+                cache: str = "",
+            ) -> None:
+                trt.IInt8Calibrator.__init__(self)
+                self.dataset = dataset
+                self.data_iter = iter(dataset)
+                self.algo = trt.CalibrationAlgoType.ENTROPY_CALIBRATION_2
+                self.batch = dataset.batch_size
+                self.cache = Path(cache)
+
+            def get_algorithm(self) -> trt.CalibrationAlgoType:
+                """Get the calibration algorithm to use."""
+                return self.algo
+
+            def get_batch_size(self) -> int:
+                """Get the batch size to use for calibration."""
+                return self.batch or 1
+
+            def get_batch(self, names) -> list:
+                """Get the next batch to use for calibration, as a list of device memory pointers."""
+                try:
+                    im0s = next(self.data_iter)["img"] / 255.0
+                    im0s = im0s.to("cuda") if im0s.device.type == "cpu" else im0s
+                    return [int(im0s.data_ptr())]
+                except StopIteration:
+                    # Return [] or None, signal to TensorRT there is no calibration data remaining
+                    return None
+
+            def read_calibration_cache(self) -> bytes:
+                """Use existing cache instead of calibrating again, otherwise, implicitly return None."""
+                if self.cache.exists() and self.cache.suffix == ".cache":
+                    return self.cache.read_bytes()
+
+            def write_calibration_cache(self, cache) -> None:
+                """Write calibration cache to disk."""
+                _ = self.cache.write_bytes(cache)
+
+        # Load dataset w/ builder (for batching) and calibrate
+        config.int8_calibrator = EngineCalibrator(
+            dataset=dataset,
+            cache=str(Path(onnx_file).with_suffix(".cache")),
+        )
+
+    elif half:
+        config.set_flag(trt.BuilderFlag.FP16)
+
+    # Write file
+    build = builder.build_serialized_network if is_trt10 else builder.build_engine
+    with build(network, config) as engine, open(engine_file, "wb") as t:
+        # Metadata
+        if metadata is not None:
+            meta = json.dumps(metadata)
+            t.write(len(meta).to_bytes(4, byteorder="little", signed=True))
+            t.write(meta.encode())
+        # Model
+        t.write(engine if is_trt10 else engine.serialize())

{ultralytics-8.3.105 → ultralytics-8.3.107}/ultralytics/utils/torch_utils.py

@@ -260,7 +260,11 @@ def fuse_conv_and_bn(conv, bn):
     fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.shape))
 
     # Prepare spatial bias
-    b_conv = torch.zeros(conv.weight.shape[0], device=conv.weight.device) if conv.bias is None else conv.bias
+    b_conv = (
+        torch.zeros(conv.weight.shape[0], dtype=conv.weight.dtype, device=conv.weight.device)
+        if conv.bias is None
+        else conv.bias
+    )
     b_bn = bn.bias - bn.weight.mul(bn.running_mean).div(torch.sqrt(bn.running_var + bn.eps))
     fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn)
 

{ultralytics-8.3.105 → ultralytics-8.3.107}/ultralytics/utils/tuner.py

@@ -137,7 +137,12 @@ def run_ray_tune(
         tuner = tune.Tuner(
             trainable_with_resources,
             param_space=space,
-            tune_config=tune.TuneConfig(scheduler=asha_scheduler, num_samples=max_samples),
+            tune_config=tune.TuneConfig(
+                scheduler=asha_scheduler,
+                num_samples=max_samples,
+                trial_name_creator=lambda trial: f"{trial.trainable_name}_{trial.trial_id}",
+                trial_dirname_creator=lambda trial: f"{trial.trainable_name}_{trial.trial_id}",
+            ),
             run_config=RunConfig(callbacks=tuner_callbacks, storage_path=tune_dir.parent, name=tune_dir.name),
         )
 

{ultralytics-8.3.105 → ultralytics-8.3.107/ultralytics.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ultralytics
-Version: 8.3.105
+Version: 8.3.107
 Summary: Ultralytics YOLO 🚀 for SOTA object detection, multi-object tracking, instance segmentation, pose estimation and image classification.
 Author-email: Glenn Jocher <glenn.jocher@ultralytics.com>, Jing Qiu <jing.qiu@ultralytics.com>
 Maintainer-email: Ultralytics <hello@ultralytics.com>
@@ -62,7 +62,7 @@ Provides-Extra: export
 Requires-Dist: onnx>=1.12.0; extra == "export"
 Requires-Dist: coremltools>=8.0; (platform_system != "Windows" and python_version <= "3.12") and extra == "export"
 Requires-Dist: scikit-learn>=1.3.2; (platform_system != "Windows" and python_version <= "3.12") and extra == "export"
-Requires-Dist: openvino!=2025.0.0,>=2024.0.0; extra == "export"
+Requires-Dist: openvino>=2024.0.0; extra == "export"
 Requires-Dist: tensorflow>=2.0.0; extra == "export"
 Requires-Dist: tensorflowjs>=4.0.0; extra == "export"
 Requires-Dist: tensorstore>=0.1.63; (platform_machine == "aarch64" and python_version >= "3.9") and extra == "export"

{ultralytics-8.3.105 → ultralytics-8.3.107}/ultralytics.egg-info/SOURCES.txt

@@ -15,15 +15,6 @@ ultralytics.egg-info/dependency_links.txt
 ultralytics.egg-info/entry_points.txt
 ultralytics.egg-info/requires.txt
 ultralytics.egg-info/top_level.txt
-ultralytics/../tests/__init__.py
-ultralytics/../tests/conftest.py
-ultralytics/../tests/test_cli.py
-ultralytics/../tests/test_cuda.py
-ultralytics/../tests/test_engine.py
-ultralytics/../tests/test_exports.py
-ultralytics/../tests/test_integrations.py
-ultralytics/../tests/test_python.py
-ultralytics/../tests/test_solutions.py
 ultralytics/assets/bus.jpg
 ultralytics/assets/zidane.jpg
 ultralytics/cfg/__init__.py
@@ -246,6 +237,7 @@ ultralytics/utils/checks.py
 ultralytics/utils/dist.py
 ultralytics/utils/downloads.py
 ultralytics/utils/errors.py
+ultralytics/utils/export.py
 ultralytics/utils/files.py
 ultralytics/utils/instance.py
 ultralytics/utils/loss.py

{ultralytics-8.3.105 → ultralytics-8.3.107}/ultralytics.egg-info/requires.txt

@@ -30,7 +30,7 @@ mkdocs-macros-plugin>=1.0.5
 
 [export]
 onnx>=1.12.0
-openvino!=2025.0.0,>=2024.0.0
+openvino>=2024.0.0
 tensorflow>=2.0.0
 tensorflowjs>=4.0.0
 keras