ultralytics 8.3.218__tar.gz → 8.3.220__tar.gz

{ultralytics-8.3.218 → ultralytics-8.3.220}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ultralytics
-Version: 8.3.218
+Version: 8.3.220
 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>

{ultralytics-8.3.218 → ultralytics-8.3.220}/tests/test_cli.py

@@ -8,7 +8,7 @@ from PIL import Image
 
 from tests import CUDA_DEVICE_COUNT, CUDA_IS_AVAILABLE, MODELS, TASK_MODEL_DATA
 from ultralytics.utils import ARM64, ASSETS, LINUX, WEIGHTS_DIR, checks
-from ultralytics.utils.torch_utils import TORCH_1_11
+from ultralytics.utils.torch_utils import TORCH_1_11, TORCH_2_9, WINDOWS
 
 
 def run(cmd: str) -> None:
@@ -129,3 +129,10 @@ def test_train_gpu(task: str, model: str, data: str) -> None:
 def test_solutions(solution: str) -> None:
     """Test yolo solutions command-line modes."""
     run(f"yolo solutions {solution} verbose=False")
+
+
+@pytest.mark.skipif(not checks.IS_PYTHON_MINIMUM_3_10 or not TORCH_2_9, reason="Requires Python>=3.10 and Torch>=2.9.0")
+@pytest.mark.skipif(WINDOWS, reason="Skipping test on Windows")
+def test_export_executorch() -> None:
+    """Test exporting a YOLO model to ExecuTorch format via CLI."""
+    run("yolo export model=yolo11n.pt format=executorch imgsz=32")

{ultralytics-8.3.218 → ultralytics-8.3.220}/tests/test_exports.py

@@ -12,15 +12,8 @@ import pytest
 from tests import MODEL, SOURCE
 from ultralytics import YOLO
 from ultralytics.cfg import TASK2DATA, TASK2MODEL, TASKS
-from ultralytics.utils import (
-    ARM64,
-    IS_RASPBERRYPI,
-    LINUX,
-    MACOS,
-    WINDOWS,
-    checks,
-)
-from ultralytics.utils.torch_utils import TORCH_1_11, TORCH_1_13, TORCH_2_1
+from ultralytics.utils import ARM64, IS_RASPBERRYPI, LINUX, MACOS, WINDOWS, checks
+from ultralytics.utils.torch_utils import TORCH_1_11, TORCH_1_13, TORCH_2_1, TORCH_2_9
 
 
 def test_export_torchscript():
@@ -262,3 +255,38 @@ def test_export_imx():
     model = YOLO("yolov8n.pt")
     file = model.export(format="imx", imgsz=32)
     YOLO(file)(SOURCE, imgsz=32)
+
+
+@pytest.mark.skipif(not checks.IS_PYTHON_MINIMUM_3_10 or not TORCH_2_9, reason="Requires Python>=3.10 and Torch>=2.9.0")
+@pytest.mark.skipif(WINDOWS, reason="Skipping test on Windows")
+def test_export_executorch():
+    """Test YOLO model export to ExecuTorch format."""
+    file = YOLO(MODEL).export(format="executorch", imgsz=32)
+    assert Path(file).exists(), f"ExecuTorch export failed, directory not found: {file}"
+    # Check that .pte file exists in the exported directory
+    pte_file = Path(file) / Path(MODEL).with_suffix(".pte").name
+    assert pte_file.exists(), f"ExecuTorch .pte file not found: {pte_file}"
+    # Check that metadata.yaml exists
+    metadata_file = Path(file) / "metadata.yaml"
+    assert metadata_file.exists(), f"ExecuTorch metadata.yaml not found: {metadata_file}"
+    # Note: Inference testing skipped as ExecuTorch requires special runtime setup
+    shutil.rmtree(file, ignore_errors=True)  # cleanup
+
+
+@pytest.mark.slow
+@pytest.mark.skipif(not checks.IS_PYTHON_MINIMUM_3_10 or not TORCH_2_9, reason="Requires Python>=3.10 and Torch>=2.9.0")
+@pytest.mark.skipif(WINDOWS, reason="Skipping test on Windows")
+@pytest.mark.parametrize("task", TASKS)
+def test_export_executorch_matrix(task):
+    """Test YOLO export to ExecuTorch format for various task types."""
+    file = YOLO(TASK2MODEL[task]).export(format="executorch", imgsz=32)
+    assert Path(file).exists(), f"ExecuTorch export failed for task '{task}', directory not found: {file}"
+    # Check that .pte file exists in the exported directory
+    model_name = Path(TASK2MODEL[task]).with_suffix(".pte").name
+    pte_file = Path(file) / model_name
+    assert pte_file.exists(), f"ExecuTorch .pte file not found for task '{task}': {pte_file}"
+    # Check that metadata.yaml exists
+    metadata_file = Path(file) / "metadata.yaml"
+    assert metadata_file.exists(), f"ExecuTorch metadata.yaml not found for task '{task}': {metadata_file}"
+    # Note: Inference testing skipped as ExecuTorch requires special runtime setup
+    shutil.rmtree(file, ignore_errors=True)  # cleanup

{ultralytics-8.3.218 → ultralytics-8.3.220}/ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 
-__version__ = "8.3.218"
+__version__ = "8.3.220"
 
 import importlib
 import os

{ultralytics-8.3.218 → ultralytics-8.3.220}/ultralytics/cfg/default.yaml

@@ -80,7 +80,7 @@ show_boxes: True # (bool) draw bounding boxes on images
 line_width: # (int, optional) line width of boxes; auto-scales with image size if not set
 
 # Export settings ------------------------------------------------------------------------------------------------------
-format: torchscript # (str) target format, e.g. torchscript|onnx|openvino|engine|coreml|saved_model|pb|tflite|edgetpu|tfjs|paddle|mnn|ncnn|imx|rknn
+format: torchscript # (str) target format, e.g. torchscript|onnx|openvino|engine|coreml|saved_model|pb|tflite|edgetpu|tfjs|paddle|mnn|ncnn|imx|rknn|executorch
 keras: False # (bool) TF SavedModel only (format=saved_model); enable Keras layers during export
 optimize: False # (bool) TorchScript only; apply mobile optimizations to the scripted model
 int8: False # (bool) INT8/PTQ where supported (openvino, tflite, tfjs, engine, imx); needs calibration data/fraction

{ultralytics-8.3.218 → ultralytics-8.3.220}/ultralytics/engine/exporter.py

@@ -20,6 +20,7 @@ MNN                     | `mnn`                     | yolo11n.mnn
 NCNN                    | `ncnn`                    | yolo11n_ncnn_model/
 IMX                     | `imx`                     | yolo11n_imx_model/
 RKNN                    | `rknn`                    | yolo11n_rknn_model/
+ExecuTorch              | `executorch`              | yolo11n_executorch_model/
 
 Requirements:
     $ pip install "ultralytics[export]"
@@ -48,6 +49,7 @@ Inference:
                          yolo11n_ncnn_model         # NCNN
                          yolo11n_imx_model          # IMX
                          yolo11n_rknn_model         # RKNN
+                         yolo11n_executorch_model   # ExecuTorch
 
 TensorFlow.js:
     $ cd .. && git clone https://github.com/zldrobit/tfjs-yolov5-example.git && cd tfjs-yolov5-example
@@ -112,7 +114,7 @@ from ultralytics.utils.metrics import batch_probiou
 from ultralytics.utils.nms import TorchNMS
 from ultralytics.utils.ops import Profile
 from ultralytics.utils.patches import arange_patch
-from ultralytics.utils.torch_utils import TORCH_1_11, TORCH_1_13, TORCH_2_1, TORCH_2_4, select_device
+from ultralytics.utils.torch_utils import TORCH_1_11, TORCH_1_13, TORCH_2_1, TORCH_2_4, TORCH_2_9, select_device
 
 
 def export_formats():
@@ -148,6 +150,7 @@ def export_formats():
         ["NCNN", "ncnn", "_ncnn_model", True, True, ["batch", "half"]],
         ["IMX", "imx", "_imx_model", True, True, ["int8", "fraction", "nms"]],
         ["RKNN", "rknn", "_rknn_model", False, False, ["batch", "name"]],
+        ["ExecuTorch", "executorch", "_executorch_model", False, False, ["batch"]],
     ]
     return dict(zip(["Format", "Argument", "Suffix", "CPU", "GPU", "Arguments"], zip(*x)))
 
@@ -322,9 +325,24 @@ class Exporter:
         flags = [x == fmt for x in fmts]
         if sum(flags) != 1:
             raise ValueError(f"Invalid export format='{fmt}'. Valid formats are {fmts}")
-        (jit, onnx, xml, engine, coreml, saved_model, pb, tflite, edgetpu, tfjs, paddle, mnn, ncnn, imx, rknn) = (
-            flags  # export booleans
-        )
+        (
+            jit,
+            onnx,
+            xml,
+            engine,
+            coreml,
+            saved_model,
+            pb,
+            tflite,
+            edgetpu,
+            tfjs,
+            paddle,
+            mnn,
+            ncnn,
+            imx,
+            rknn,
+            executorch,
+        ) = flags  # export booleans
 
         is_tf_format = any((saved_model, pb, tflite, edgetpu, tfjs))
 
@@ -543,6 +561,8 @@ class Exporter:
             f[13] = self.export_imx()
         if rknn:
             f[14] = self.export_rknn()
+        if executorch:
+            f[15] = self.export_executorch()
 
         # Finish
         f = [str(x) for x in f if x]  # filter out '' and None
@@ -1122,6 +1142,39 @@ class Exporter:
             f = saved_model / f"{self.file.stem}_float32.tflite"
         return str(f)
 
+    @try_export
+    def export_executorch(self, prefix=colorstr("ExecuTorch:")):
+        """Exports a model to ExecuTorch (.pte) format into a dedicated directory and saves the required metadata,
+        following Ultralytics conventions.
+        """
+        LOGGER.info(f"\n{prefix} starting export with ExecuTorch...")
+        assert TORCH_2_9, f"ExecuTorch export requires torch>=2.9.0 but torch=={TORCH_VERSION} is installed"
+        # TorchAO release compatibility table bug https://github.com/pytorch/ao/issues/2919
+        # Setuptools bug: https://github.com/pypa/setuptools/issues/4483
+        check_requirements("setuptools<71.0.0")  # Setuptools bug: https://github.com/pypa/setuptools/issues/4483
+        check_requirements(("executorch==1.0.0", "flatbuffers"))
+
+        import torch
+        from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
+        from executorch.exir import to_edge_transform_and_lower
+
+        file_directory = Path(str(self.file).replace(self.file.suffix, "_executorch_model"))
+        file_directory.mkdir(parents=True, exist_ok=True)
+
+        file_pte = file_directory / self.file.with_suffix(".pte").name
+        sample_inputs = (self.im,)
+
+        et_program = to_edge_transform_and_lower(
+            torch.export.export(self.model, sample_inputs), partitioner=[XnnpackPartitioner()]
+        ).to_executorch()
+
+        with open(file_pte, "wb") as file:
+            file.write(et_program.buffer)
+
+        YAML.save(file_directory / "metadata.yaml", self.metadata)
+
+        return str(file_directory)
+
     @try_export
     def export_edgetpu(self, tflite_model="", prefix=colorstr("Edge TPU:")):
         """Export YOLO model to Edge TPU format https://coral.ai/docs/edgetpu/models-intro/."""

{ultralytics-8.3.218 → ultralytics-8.3.220}/ultralytics/engine/predictor.py

@@ -30,6 +30,7 @@ Usage - formats:
                               yolo11n_ncnn_model         # NCNN
                               yolo11n_imx_model          # Sony IMX
                               yolo11n_rknn_model         # Rockchip RKNN
+                              yolo11n.pte                # PyTorch Executorch
 """
 
 from __future__ import annotations

{ultralytics-8.3.218 → ultralytics-8.3.220}/ultralytics/models/fastsam/predict.py

@@ -4,7 +4,7 @@ import torch
 from PIL import Image
 
 from ultralytics.models.yolo.segment import SegmentationPredictor
-from ultralytics.utils import DEFAULT_CFG, checks
+from ultralytics.utils import DEFAULT_CFG
 from ultralytics.utils.metrics import box_iou
 from ultralytics.utils.ops import scale_masks
 from ultralytics.utils.torch_utils import TORCH_1_10
@@ -101,7 +101,7 @@ class FastSAMPredictor(SegmentationPredictor):
                 continue
             masks = result.masks.data
             if masks.shape[1:] != result.orig_shape:
-                masks = scale_masks(masks[None], result.orig_shape)[0]
+                masks = (scale_masks(masks[None].float(), result.orig_shape)[0] > 0.5).byte()
             # bboxes prompt
             idx = torch.zeros(len(result), dtype=torch.bool, device=self.device)
             if bboxes is not None:
@@ -161,20 +161,14 @@ class FastSAMPredictor(SegmentationPredictor):
         Returns:
             (torch.Tensor): Similarity matrix between given images and texts with shape (M, N).
         """
-        try:
-            import clip
-        except ImportError:
-            checks.check_requirements("git+https://github.com/ultralytics/CLIP.git")
-            import clip
-        if (not hasattr(self, "clip_model")) or (not hasattr(self, "clip_preprocess")):
-            self.clip_model, self.clip_preprocess = clip.load("ViT-B/32", device=self.device)
-        images = torch.stack([self.clip_preprocess(image).to(self.device) for image in images])
-        tokenized_text = clip.tokenize(texts).to(self.device)
-        image_features = self.clip_model.encode_image(images)
-        text_features = self.clip_model.encode_text(tokenized_text)
-        image_features /= image_features.norm(dim=-1, keepdim=True)  # (N, 512)
-        text_features /= text_features.norm(dim=-1, keepdim=True)  # (M, 512)
-        return (image_features * text_features[:, None]).sum(-1)  # (M, N)
+        from ultralytics.nn.text_model import CLIP
+
+        if not hasattr(self, "clip"):
+            self.clip = CLIP("ViT-B/32", device=self.device)
+        images = torch.stack([self.clip.image_preprocess(image).to(self.device) for image in images])
+        image_features = self.clip.encode_image(images)
+        text_features = self.clip.encode_text(self.clip.tokenize(texts))
+        return text_features @ image_features.T  # (M, N)
 
     def set_prompts(self, prompts):
         """Set prompts to be used during inference."""

{ultralytics-8.3.218 → ultralytics-8.3.220}/ultralytics/nn/autobackend.py

@@ -96,6 +96,8 @@ class AutoBackend(nn.Module):
             | NCNN                  | *_ncnn_model/     |
             | IMX                   | *_imx_model/      |
             | RKNN                  | *_rknn_model/     |
+            | Triton Inference      | triton://model    |
+            | ExecuTorch            | *.pte             |
 
     Attributes:
         model (torch.nn.Module): The loaded YOLO model.
@@ -122,6 +124,7 @@ class AutoBackend(nn.Module):
         imx (bool): Whether the model is an IMX model.
         rknn (bool): Whether the model is an RKNN model.
         triton (bool): Whether the model is a Triton Inference Server model.
+        pte (bool): Whether the model is a PyTorch ExecuTorch model.
 
     Methods:
         forward: Run inference on an input image.
@@ -176,6 +179,7 @@ class AutoBackend(nn.Module):
             ncnn,
             imx,
             rknn,
+            pte,
             triton,
         ) = self._model_type("" if nn_module else model)
         fp16 &= pt or jit or onnx or xml or engine or nn_module or triton  # FP16
@@ -570,6 +574,25 @@ class AutoBackend(nn.Module):
             rknn_model.init_runtime()
             metadata = w.parent / "metadata.yaml"
 
+        # ExecuTorch
+        elif pte:
+            LOGGER.info(f"Loading {w} for ExecuTorch inference...")
+            # TorchAO release compatibility table bug https://github.com/pytorch/ao/issues/2919
+            check_requirements("setuptools<71.0.0")  # Setuptools bug: https://github.com/pypa/setuptools/issues/4483
+            check_requirements(("executorch==1.0.0", "flatbuffers"))
+            from executorch.runtime import Runtime
+
+            w = Path(w)
+            if w.is_dir():
+                model_file = next(w.rglob("*.pte"))
+                metadata = w / "metadata.yaml"
+            else:
+                model_file = w
+                metadata = w.parent / "metadata.yaml"
+
+            program = Runtime.get().load_program(str(model_file))
+            model = program.load_method("forward")
+
         # Any other format (unsupported)
         else:
             from ultralytics.engine.exporter import export_formats
@@ -773,6 +796,10 @@ class AutoBackend(nn.Module):
             im = im if isinstance(im, (list, tuple)) else [im]
             y = self.rknn_model.inference(inputs=im)
 
+        # ExecuTorch
+        elif self.pte:
+            y = self.model.execute([im])
+
         # TensorFlow (SavedModel, GraphDef, Lite, Edge TPU)
         else:
             im = im.cpu().numpy()

{ultralytics-8.3.218 → ultralytics-8.3.220}/ultralytics/solutions/similarity_search.py

@@ -78,11 +78,11 @@ class VisualAISearch:
 
     def extract_image_feature(self, path: Path) -> np.ndarray:
         """Extract CLIP image embedding from the given image path."""
-        return self.model.encode_image(Image.open(path)).cpu().numpy()
+        return self.model.encode_image(Image.open(path)).detach().cpu().numpy()
 
     def extract_text_feature(self, text: str) -> np.ndarray:
         """Extract CLIP text embedding from the given text query."""
-        return self.model.encode_text(self.model.tokenize([text])).cpu().numpy()
+        return self.model.encode_text(self.model.tokenize([text])).detach().cpu().numpy()
 
     def load_or_build_index(self) -> None:
         """

{ultralytics-8.3.218 → ultralytics-8.3.220}/ultralytics/utils/torch_utils.py

@@ -44,6 +44,7 @@ TORCH_1_13 = check_version(TORCH_VERSION, "1.13.0")
 TORCH_2_0 = check_version(TORCH_VERSION, "2.0.0")
 TORCH_2_1 = check_version(TORCH_VERSION, "2.1.0")
 TORCH_2_4 = check_version(TORCH_VERSION, "2.4.0")
+TORCH_2_9 = check_version(TORCH_VERSION, "2.9.0")
 TORCHVISION_0_10 = check_version(TORCHVISION_VERSION, "0.10.0")
 TORCHVISION_0_11 = check_version(TORCHVISION_VERSION, "0.11.0")
 TORCHVISION_0_13 = check_version(TORCHVISION_VERSION, "0.13.0")

{ultralytics-8.3.218 → ultralytics-8.3.220}/ultralytics.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ultralytics
-Version: 8.3.218
+Version: 8.3.220
 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>