`ultralytics 8.1.15` add Python 3.12 compatibility (#8210)

Signed-off-by: Glenn Jocher <glenn.jocher@ultralytics.com>
Co-authored-by: Johnny <johnnynuca14@gmail.com>
Co-authored-by: UltralyticsAssistant <web@ultralytics.com>

.github/workflows/ci.yaml

@@ -179,7 +179,7 @@ jobs:
           if [ "${{ matrix.torch }}" == "1.8.0" ]; then
               torch="torch==1.8.0 torchvision==0.9.0"
           fi
-          pip install -e . $torch pytest-cov "coremltools>=7.0" --extra-index-url https://download.pytorch.org/whl/cpu
+          pip install -e . $torch pytest-cov "coremltools>=7.0; platform_system != 'Windows' and python_version <= '3.11'" --extra-index-url https://download.pytorch.org/whl/cpu
       - name: Check environment
         run: |
           yolo checks
@@ -249,7 +249,8 @@ jobs:
           conda install -c pytorch -c conda-forge pytorch torchvision ultralytics openvino
       - name: Install pip packages
         run: |
-          pip install pytest 'coremltools>=7.0'
+          # CoreML must be installed before export due to protobuf error from AutoInstall
+          pip install pytest "coremltools>=7.0; platform_system != 'Windows' and python_version <= '3.11'"
       - name: Check environment
         run: |
           conda list

pyproject.toml

@@ -52,6 +52,7 @@ classifiers = [
     "Programming Language :: Python :: 3.9",
     "Programming Language :: Python :: 3.10",
     "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
     "Topic :: Software Development",
     "Topic :: Scientific/Engineering",
     "Topic :: Scientific/Engineering :: Artificial Intelligence",
@@ -64,7 +65,6 @@ classifiers = [
 # Required dependencies ------------------------------------------------------------------------------------------------
 dependencies = [
     "matplotlib>=3.3.0",
-    "numpy>=1.22.2",
     "opencv-python>=4.6.0",
     "pillow>=7.1.2",
     "pyyaml>=5.3.1",
@@ -97,10 +97,10 @@ dev = [
 ]
 export = [
     "onnx>=1.12.0", # ONNX export
-    "coremltools>=7.0; platform_system != 'Windows'", # CoreML only supported on macOS and Linux
-    "openvino-dev>=2023.0", # OpenVINO export
-    "tensorflow<=2.13.1", # TF bug https://github.com/ultralytics/ultralytics/issues/5161
-    "tensorflowjs>=3.9.0", # TF.js export, automatically installs tensorflow
+    "coremltools>=7.0; platform_system != 'Windows' and python_version <= '3.11'", # CoreML supported on macOS and Linux
+    "openvino-dev>=2023.0; python_version <= '3.11'", # OpenVINO export
+    "tensorflow<=2.13.1; python_version <= '3.11'", # TF bug https://github.com/ultralytics/ultralytics/issues/5161
+    "tensorflowjs>=3.9.0; python_version <= '3.11'", # TF.js export, automatically installs tensorflow
 ]
 explorer = [
     "lancedb", # vector search
@@ -120,7 +120,7 @@ extra = [
     "hub-sdk>=0.0.2", # Ultralytics HUB
     "ipython", # interactive notebook
     "albumentations>=1.0.3", # training augmentations
-    "pycocotools>=2.0.6", # COCO mAP
+    "pycocotools>=2.0.7", # COCO mAP
 ]
 
 [project.urls]

tests/test_cli.py

@@ -4,11 +4,10 @@ import subprocess
 
 import pytest
 
-from ultralytics.utils import ASSETS, WEIGHTS_DIR
-from ultralytics.utils.checks import cuda_device_count, cuda_is_available
+from ultralytics.utils import ASSETS, WEIGHTS_DIR, checks
 
-CUDA_IS_AVAILABLE = cuda_is_available()
-CUDA_DEVICE_COUNT = cuda_device_count()
+CUDA_IS_AVAILABLE = checks.cuda_is_available()
+CUDA_DEVICE_COUNT = checks.cuda_device_count()
 TASK_ARGS = [
     ("detect", "yolov8n", "coco8.yaml"),
     ("segment", "yolov8n-seg", "coco8-seg.yaml"),
@@ -70,6 +69,7 @@ def test_rtdetr(task="detect", model="yolov8n-rtdetr.yaml", data="coco8.yaml"):
     run(f"yolo predict {task} model={model} source={ASSETS / 'bus.jpg'} imgsz=640 save save_crop save_txt")
 
 
+@pytest.mark.skipif(checks.IS_PYTHON_3_12, reason="MobileSAM Clip is not supported in Python 3.12")
 def test_fastsam(task="segment", model=WEIGHTS_DIR / "FastSAM-s.pt", data="coco8-seg.yaml"):
     """Test FastSAM segmentation functionality within Ultralytics."""
     source = ASSETS / "bus.jpg"

tests/test_integrations.py

@@ -41,16 +41,16 @@ def test_triton():
 
     # Create variables
     model_name = "yolo"
-    triton_repo_path = TMP / "triton_repo"
-    triton_model_path = triton_repo_path / model_name
+    triton_repo = TMP / "triton_repo"  # Triton repo path
+    triton_model = triton_repo / model_name  # Triton model path
 
     # Export model to ONNX
     f = YOLO(MODEL).export(format="onnx", dynamic=True)
 
     # Prepare Triton repo
-    (triton_model_path / "1").mkdir(parents=True, exist_ok=True)
-    Path(f).rename(triton_model_path / "1" / "model.onnx")
-    (triton_model_path / "config.pbtxt").touch()
+    (triton_model / "1").mkdir(parents=True, exist_ok=True)
+    Path(f).rename(triton_model / "1" / "model.onnx")
+    (triton_model / "config.pbtxt").touch()
 
     # Define image https://catalog.ngc.nvidia.com/orgs/nvidia/containers/tritonserver
     tag = "nvcr.io/nvidia/tritonserver:23.09-py3"  # 6.4 GB
@@ -61,7 +61,7 @@ def test_triton():
     # Run the Triton server and capture the container ID
     container_id = (
         subprocess.check_output(
-            f"docker run -d --rm -v {triton_repo_path}:/models -p 8000:8000 {tag} tritonserver --model-repository=/models",
+            f"docker run -d --rm -v {triton_repo}:/models -p 8000:8000 {tag} tritonserver --model-repository=/models",
             shell=True,
         )
         .decode("utf-8")

tests/test_python.py

@@ -218,12 +218,14 @@ def test_export_onnx():
     YOLO(f)(SOURCE)  # exported model inference
 
 
+@pytest.mark.skipif(checks.IS_PYTHON_3_12, reason="OpenVINO not supported in Python 3.12")
 def test_export_openvino():
     """Test exporting the YOLO model to OpenVINO format."""
     f = YOLO(MODEL).export(format="openvino")
     YOLO(f)(SOURCE)  # exported model inference
 
 
+@pytest.mark.skipif(checks.IS_PYTHON_3_12, reason="CoreML not supported in Python 3.12")
 def test_export_coreml():
     """Test exporting the YOLO model to CoreML format."""
     if not WINDOWS:  # RuntimeError: BlobWriter not loaded with coremltools 7.0 on windows

ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
 
-__version__ = "8.1.14"
+__version__ = "8.1.15"
 
 from ultralytics.data.explorer.explorer import Explorer
 from ultralytics.models import RTDETR, SAM, YOLO, YOLOWorld

ultralytics/engine/exporter.py

@@ -81,7 +81,7 @@ from ultralytics.utils import (
     get_default_args,
     yaml_save,
 )
-from ultralytics.utils.checks import check_imgsz, check_is_path_safe, check_requirements, check_version
+from ultralytics.utils.checks import PYTHON_VERSION, check_imgsz, check_is_path_safe, check_requirements, check_version
 from ultralytics.utils.downloads import attempt_download_asset, get_github_assets
 from ultralytics.utils.files import file_size, spaces_in_path
 from ultralytics.utils.ops import Profile
@@ -609,10 +609,8 @@ class Exporter:
                 ct_model = cto.palettize_weights(ct_model, config=config)
         if self.args.nms and self.model.task == "detect":
             if mlmodel:
-                import platform
-
                 # coremltools<=6.2 NMS export requires Python<3.11
-                check_version(platform.python_version(), "<3.11", name="Python ", hard=True)
+                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

ultralytics/engine/trainer.py

@@ -285,7 +285,7 @@ class BaseTrainer:
         batch_size = self.batch_size // max(world_size, 1)
         self.train_loader = self.get_dataloader(self.trainset, batch_size=batch_size, rank=RANK, mode="train")
         if RANK in (-1, 0):
-            # NOTE: When training DOTA dataset, double batch size could get OOM cause some images got more than 2000 objects.
+            # Note: When training DOTA dataset, double batch size could get OOM on images with >2000 objects.
             self.test_loader = self.get_dataloader(
                 self.testset, batch_size=batch_size if self.args.task == "obb" else batch_size * 2, rank=-1, mode="val"
             )

ultralytics/utils/benchmarks.py

@@ -36,7 +36,7 @@ from ultralytics import YOLO
 from ultralytics.cfg import TASK2DATA, TASK2METRIC
 from ultralytics.engine.exporter import export_formats
 from ultralytics.utils import ASSETS, LINUX, LOGGER, MACOS, TQDM, WEIGHTS_DIR
-from ultralytics.utils.checks import check_requirements, check_yolo
+from ultralytics.utils.checks import IS_PYTHON_3_12, check_requirements, check_yolo
 from ultralytics.utils.files import file_size
 from ultralytics.utils.torch_utils import select_device
 
@@ -90,6 +90,8 @@ def benchmark(
                 assert model.task != "obb", "TensorFlow GraphDef not supported for OBB task"
             elif i in {5, 10}:  # CoreML and TF.js
                 assert MACOS or LINUX, "export only supported on macOS and Linux"
+            if i in {3, 5}:  # CoreML and OpenVINO
+                assert not IS_PYTHON_3_12, "CoreML and OpenVINO not supported on Python 3.12"
             if "cpu" in device.type:
                 assert cpu, "inference not supported on CPU"
             if "cuda" in device.type:
@@ -147,8 +149,7 @@ class ProfileModels:
     """
     ProfileModels class for profiling different models on ONNX and TensorRT.
 
-    This class profiles the performance of different models, provided their paths. The profiling includes parameters such as
-    model speed and FLOPs.
+    This class profiles the performance of different models, returning results such as model speed and FLOPs.
 
     Attributes:
         paths (list): Paths of the models to profile.
@@ -188,9 +189,9 @@ class ProfileModels:
             num_warmup_runs (int, optional): Number of warmup runs before the actual profiling starts. Default is 10.
             min_time (float, optional): Minimum time in seconds for profiling a model. Default is 60.
             imgsz (int, optional): Size of the image used during profiling. Default is 640.
-            half (bool, optional): Flag to indicate whether to use half-precision floating point for profiling. Default is True.
+            half (bool, optional): Flag to indicate whether to use half-precision floating point for profiling.
             trt (bool, optional): Flag to indicate whether to profile using TensorRT. Default is True.
-            device (torch.device, optional): Device used for profiling. If None, it is determined automatically. Default is None.
+            device (torch.device, optional): Device used for profiling. If None, it is determined automatically.
         """
         self.paths = paths
         self.num_timed_runs = num_timed_runs

ultralytics/utils/callbacks/wb.py

@@ -23,8 +23,9 @@ def _custom_table(x, y, classes, title="Precision Recall Curve", x_title="Recall
     """
     Create and log a custom metric visualization to wandb.plot.pr_curve.
 
-    This function crafts a custom metric visualization that mimics the behavior of wandb's default precision-recall curve
-    while allowing for enhanced customization. The visual metric is useful for monitoring model performance across different classes.
+    This function crafts a custom metric visualization that mimics the behavior of wandb's default precision-recall
+    curve while allowing for enhanced customization. The visual metric is useful for monitoring model performance across
+    different classes.
 
     Args:
         x (List): Values for the x-axis; expected to have length N.
@@ -64,8 +65,8 @@ def _plot_curve(
 
     Args:
         x (np.ndarray): Data points for the x-axis with length N.
-        y (np.ndarray): Corresponding data points for the y-axis with shape CxN, where C represents the number of classes.
-        names (list, optional): Names of the classes corresponding to the y-axis data; length C. Defaults to an empty list.
+        y (np.ndarray): Corresponding data points for the y-axis with shape CxN, where C is the number of classes.
+        names (list, optional): Names of the classes corresponding to the y-axis data; length C. Defaults to [].
         id (str, optional): Unique identifier for the logged data in wandb. Defaults to 'precision-recall'.
         title (str, optional): Title for the visualization plot. Defaults to 'Precision Recall Curve'.
         x_title (str, optional): Label for the x-axis. Defaults to 'Recall'.

ultralytics/utils/checks.py

@@ -9,7 +9,6 @@ import platform
 import re
 import shutil
 import subprocess
-import sys
 import time
 from importlib import metadata
 from pathlib import Path
@@ -46,6 +45,8 @@ from ultralytics.utils import (
     url2file,
 )
 
+PYTHON_VERSION = platform.python_version()
+
 
 def parse_requirements(file_path=ROOT.parent / "requirements.txt", package=""):
     """
@@ -329,7 +330,7 @@ def check_python(minimum: str = "3.8.0") -> bool:
     Returns:
         (bool): Whether the installed Python version meets the minimum constraints.
     """
-    return check_version(platform.python_version(), minimum, name="Python ", hard=True)
+    return check_version(PYTHON_VERSION, minimum, name="Python ", hard=True)
 
 
 @TryExcept()
@@ -580,7 +581,7 @@ def collect_system_info():
     LOGGER.info(
         f"\n{'OS':<20}{platform.platform()}\n"
         f"{'Environment':<20}{ENVIRONMENT}\n"
-        f"{'Python':<20}{sys.version.split()[0]}\n"
+        f"{'Python':<20}{PYTHON_VERSION}\n"
         f"{'Install':<20}{'git' if is_git_dir() else 'pip' if is_pip_package() else 'other'}\n"
         f"{'RAM':<20}{ram_info:.2f} GB\n"
         f"{'CPU':<20}{get_cpu_info()}\n"
@@ -722,3 +723,7 @@ def cuda_is_available() -> bool:
         (bool): True if one or more NVIDIA GPUs are available, False otherwise.
     """
     return cuda_device_count() > 0
+
+
+# Define constants
+IS_PYTHON_3_12 = check_version(PYTHON_VERSION, "==3.12", name="Python ", hard=False)

ultralytics/utils/ops.py

@@ -546,7 +546,7 @@ def xywhr2xyxyxyxy(rboxes):
     be in degrees from 0 to 90.
 
     Args:
-        rboxes (numpy.ndarray | torch.Tensor): Input data in [cx, cy, w, h, rotation] format of shape (n, 5) or (b, n, 5).
+        rboxes (numpy.ndarray | torch.Tensor): Boxes in [cx, cy, w, h, rotation] format of shape (n, 5) or (b, n, 5).
 
     Returns:
         (numpy.ndarray | torch.Tensor): Converted corner points of shape (n, 4, 2) or (b, n, 4, 2).

ultralytics/utils/torch_utils.py

@@ -2,7 +2,6 @@
 
 import math
 import os
-import platform
 import random
 import time
 from contextlib import contextmanager
@@ -18,7 +17,7 @@ import torch.nn.functional as F
 import torchvision
 
 from ultralytics.utils import DEFAULT_CFG_DICT, DEFAULT_CFG_KEYS, LOGGER, __version__
-from ultralytics.utils.checks import check_version
+from ultralytics.utils.checks import PYTHON_VERSION, check_version
 
 try:
     import thop
@@ -103,7 +102,7 @@ def select_device(device="", batch=0, newline=False, verbose=True):
     if isinstance(device, torch.device):
         return device
 
-    s = f"Ultralytics YOLOv{__version__} 🚀 Python-{platform.python_version()} torch-{torch.__version__} "
+    s = f"Ultralytics YOLOv{__version__} 🚀 Python-{PYTHON_VERSION} torch-{torch.__version__} "
     device = str(device).lower()
     for remove in "cuda:", "none", "(", ")", "[", "]", "'", " ":
         device = device.replace(remove, "")  # to string, 'cuda:0' -> '0' and '(0, 1)' -> '0,1'