ultralytics 8.3.28__py3-none-any.whl → 8.3.29__py3-none-any.whl

tests/test_exports.py

@@ -205,3 +205,12 @@ def test_export_ncnn():
     """Test YOLO exports to NCNN format."""
     file = YOLO(MODEL).export(format="ncnn", imgsz=32)
     YOLO(file)(SOURCE, imgsz=32)  # exported model inference
+
+
+@pytest.mark.skipif(True, reason="Test disabled as keras and tensorflow version conflicts with tflite export.")
+@pytest.mark.skipif(not LINUX or MACOS, reason="Skipping test on Windows and Macos")
+def test_export_imx():
+    """Test YOLOv8n exports to IMX format."""
+    model = YOLO("yolov8n.pt")
+    file = model.export(format="imx", imgsz=32)
+    YOLO(file)(SOURCE, imgsz=32)

ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics YOLO 🚀, AGPL-3.0 license
 
-__version__ = "8.3.28"
+__version__ = "8.3.29"
 
 import os
 

ultralytics/data/converter.py

@@ -577,7 +577,7 @@ def merge_multi_segment(segments):
     return s
 
 
-def yolo_bbox2segment(im_dir, save_dir=None, sam_model="sam_b.pt"):
+def yolo_bbox2segment(im_dir, save_dir=None, sam_model="sam_b.pt", device=None):
     """
     Converts existing object detection dataset (bounding boxes) to segmentation dataset or oriented bounding box (OBB)
     in YOLO format. Generates segmentation data using SAM auto-annotator as needed.
@@ -587,6 +587,7 @@ def yolo_bbox2segment(im_dir, save_dir=None, sam_model="sam_b.pt"):
         save_dir (str | Path): Path to save the generated labels, labels will be saved
             into `labels-segment` in the same directory level of `im_dir` if save_dir is None. Default: None.
         sam_model (str): Segmentation model to use for intermediate segmentation data; optional.
+        device (int | str): The specific device to run SAM models. Default: None.
 
     Notes:
         The input directory structure assumed for dataset:
@@ -621,7 +622,7 @@ def yolo_bbox2segment(im_dir, save_dir=None, sam_model="sam_b.pt"):
         boxes[:, [0, 2]] *= w
         boxes[:, [1, 3]] *= h
         im = cv2.imread(label["im_file"])
-        sam_results = sam_model(im, bboxes=xywh2xyxy(boxes), verbose=False, save=False)
+        sam_results = sam_model(im, bboxes=xywh2xyxy(boxes), verbose=False, save=False, device=device)
         label["segments"] = sam_results[0].masks.xyn
 
     save_dir = Path(save_dir) if save_dir else Path(im_dir).parent / "labels-segment"
@@ -636,8 +637,8 @@ def yolo_bbox2segment(im_dir, save_dir=None, sam_model="sam_b.pt"):
                 continue
             line = (int(cls[i]), *s.reshape(-1))
             texts.append(("%g " * len(line)).rstrip() % line)
-            with open(txt_file, "a") as f:
-                f.writelines(text + "\n" for text in texts)
+        with open(txt_file, "a") as f:
+            f.writelines(text + "\n" for text in texts)
     LOGGER.info(f"Generated segment labels saved in {save_dir}")
 
 

ultralytics/engine/exporter.py

@@ -18,6 +18,7 @@ TensorFlow.js           | `tfjs`                    | yolo11n_web_model/
 PaddlePaddle            | `paddle`                  | yolo11n_paddle_model/
 MNN                     | `mnn`                     | yolo11n.mnn
 NCNN                    | `ncnn`                    | yolo11n_ncnn_model/
+IMX                     | `imx`                     | yolo11n_imx_model/
 
 Requirements:
     $ pip install "ultralytics[export]"
@@ -44,6 +45,7 @@ Inference:
                          yolo11n_paddle_model       # PaddlePaddle
                          yolo11n.mnn                # MNN
                          yolo11n_ncnn_model         # NCNN
+                         yolo11n_imx_model          # IMX
 
 TensorFlow.js:
     $ cd .. && git clone https://github.com/zldrobit/tfjs-yolov5-example.git && cd tfjs-yolov5-example
@@ -94,7 +96,7 @@ from ultralytics.utils.checks import check_imgsz, check_is_path_safe, check_requ
 from ultralytics.utils.downloads import attempt_download_asset, get_github_assets, safe_download
 from ultralytics.utils.files import file_size, spaces_in_path
 from ultralytics.utils.ops import Profile
-from ultralytics.utils.torch_utils import TORCH_1_13, get_latest_opset, select_device, smart_inference_mode
+from ultralytics.utils.torch_utils import TORCH_1_13, get_latest_opset, select_device
 
 
 def export_formats():
@@ -114,6 +116,7 @@ def export_formats():
         ["PaddlePaddle", "paddle", "_paddle_model", True, True],
         ["MNN", "mnn", ".mnn", True, True],
         ["NCNN", "ncnn", "_ncnn_model", True, True],
+        ["IMX", "imx", "_imx_model", True, True],
     ]
     return dict(zip(["Format", "Argument", "Suffix", "CPU", "GPU"], zip(*x)))
 
@@ -171,7 +174,6 @@ class Exporter:
         self.callbacks = _callbacks or callbacks.get_default_callbacks()
         callbacks.add_integration_callbacks(self)
 
-    @smart_inference_mode()
     def __call__(self, model=None) -> str:
         """Returns list of exported files/dirs after running callbacks."""
         self.run_callbacks("on_export_start")
@@ -194,9 +196,22 @@ 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 = (
-            flags  # export booleans
-        )
+        (
+            jit,
+            onnx,
+            xml,
+            engine,
+            coreml,
+            saved_model,
+            pb,
+            tflite,
+            edgetpu,
+            tfjs,
+            paddle,
+            mnn,
+            ncnn,
+            imx,
+        ) = flags  # export booleans
         is_tf_format = any((saved_model, pb, tflite, edgetpu, tfjs))
 
         # Device
@@ -210,6 +225,9 @@ class Exporter:
         self.device = select_device("cpu" if self.args.device is None else self.args.device)
 
         # Checks
+        if imx and not self.args.int8:
+            LOGGER.warning("WARNING ⚠️ IMX only supports int8 export, setting int8=True.")
+            self.args.int8 = True
         if not hasattr(model, "names"):
             model.names = default_class_names()
         model.names = check_class_names(model.names)
@@ -227,7 +245,7 @@ class Exporter:
             assert not ncnn, "optimize=True not compatible with format='ncnn', i.e. use optimize=False"
             assert self.device.type == "cpu", "optimize=True not compatible with cuda devices, i.e. use device='cpu'"
         if self.args.int8 and tflite:
-            assert not model.end2end, "TFLite INT8 export not supported for end2end models, please use half precision."
+            assert not getattr(model, "end2end", False), "TFLite INT8 export not supported for end2end models."
         if edgetpu:
             if not LINUX:
                 raise SystemError("Edge TPU export only supported on Linux. See https://coral.ai/docs/edgetpu/compiler")
@@ -249,6 +267,7 @@ class Exporter:
             )
         if mnn and (IS_RASPBERRYPI or IS_JETSON):
             raise SystemError("MNN export not supported on Raspberry Pi and NVIDIA Jetson")
+
         # Input
         im = torch.zeros(self.args.batch, 3, *self.imgsz).to(self.device)
         file = Path(
@@ -264,6 +283,11 @@ class Exporter:
         model.eval()
         model.float()
         model = model.fuse()
+
+        if imx:
+            from ultralytics.utils.torch_utils import FXModel
+
+            model = FXModel(model)
         for m in model.modules():
             if isinstance(m, (Detect, RTDETRDecoder)):  # includes all Detect subclasses like Segment, Pose, OBB
                 m.dynamic = self.args.dynamic
@@ -273,6 +297,15 @@ class Exporter:
             elif isinstance(m, C2f) and not is_tf_format:
                 # EdgeTPU does not support FlexSplitV while split provides cleaner ONNX graph
                 m.forward = m.forward_split
+            if isinstance(m, Detect) and imx:
+                from ultralytics.utils.tal import make_anchors
+
+                m.anchors, m.strides = (
+                    x.transpose(0, 1)
+                    for x in make_anchors(
+                        torch.cat([s / m.stride.unsqueeze(-1) for s in self.imgsz], dim=1), m.stride, 0.5
+                    )
+                )
 
         y = None
         for _ in range(2):
@@ -347,6 +380,8 @@ class Exporter:
             f[11], _ = self.export_mnn()
         if ncnn:  # NCNN
             f[12], _ = self.export_ncnn()
+        if imx:
+            f[13], _ = self.export_imx()
 
         # Finish
         f = [str(x) for x in f if x]  # filter out '' and None
@@ -568,8 +603,7 @@ class Exporter:
         f = str(self.file.with_suffix(".mnn"))  # MNN model file
         args = ["", "-f", "ONNX", "--modelFile", f_onnx, "--MNNModel", f, "--bizCode", json.dumps(self.metadata)]
         if self.args.int8:
-            args.append("--weightQuantBits")
-            args.append("8")
+            args.extend(("--weightQuantBits", "8"))
         if self.args.half:
             args.append("--fp16")
         mnnconvert.convert(args)
@@ -1069,6 +1103,137 @@ class Exporter:
         yaml_save(Path(f) / "metadata.yaml", self.metadata)  # add metadata.yaml
         return f, None
 
+    @try_export
+    def export_imx(self, prefix=colorstr("IMX:")):
+        """YOLO IMX export."""
+        gptq = False
+        assert LINUX, "export only supported on Linux. See https://developer.aitrios.sony-semicon.com/en/raspberrypi-ai-camera/documentation/imx500-converter"
+        if getattr(self.model, "end2end", False):
+            raise ValueError("IMX export is not supported for end2end models.")
+        if "C2f" not in self.model.__str__():
+            raise ValueError("IMX export is only supported for YOLOv8 detection models")
+        check_requirements(("model-compression-toolkit==2.1.1", "sony-custom-layers==0.2.0", "tensorflow==2.12.0"))
+        check_requirements("imx500-converter[pt]==3.14.3")  # Separate requirements for imx500-converter
+
+        import model_compression_toolkit as mct
+        import onnx
+        from sony_custom_layers.pytorch.object_detection.nms import multiclass_nms
+
+        try:
+            out = subprocess.run(
+                ["java", "--version"], check=True, capture_output=True
+            )  # Java 17 is required for imx500-converter
+            if "openjdk 17" not in str(out.stdout):
+                raise FileNotFoundError
+        except FileNotFoundError:
+            subprocess.run(["sudo", "apt", "install", "-y", "openjdk-17-jdk", "openjdk-17-jre"], check=True)
+
+        def representative_dataset_gen(dataloader=self.get_int8_calibration_dataloader(prefix)):
+            for batch in dataloader:
+                img = batch["img"]
+                img = img / 255.0
+                yield [img]
+
+        tpc = mct.get_target_platform_capabilities(
+            fw_name="pytorch", target_platform_name="imx500", target_platform_version="v1"
+        )
+
+        config = mct.core.CoreConfig(
+            mixed_precision_config=mct.core.MixedPrecisionQuantizationConfig(num_of_images=10),
+            quantization_config=mct.core.QuantizationConfig(concat_threshold_update=True),
+        )
+
+        resource_utilization = mct.core.ResourceUtilization(weights_memory=3146176 * 0.76)
+
+        quant_model = (
+            mct.gptq.pytorch_gradient_post_training_quantization(  # Perform Gradient-Based Post Training Quantization
+                model=self.model,
+                representative_data_gen=representative_dataset_gen,
+                target_resource_utilization=resource_utilization,
+                gptq_config=mct.gptq.get_pytorch_gptq_config(n_epochs=1000, use_hessian_based_weights=False),
+                core_config=config,
+                target_platform_capabilities=tpc,
+            )[0]
+            if gptq
+            else mct.ptq.pytorch_post_training_quantization(  # Perform post training quantization
+                in_module=self.model,
+                representative_data_gen=representative_dataset_gen,
+                target_resource_utilization=resource_utilization,
+                core_config=config,
+                target_platform_capabilities=tpc,
+            )[0]
+        )
+
+        class NMSWrapper(torch.nn.Module):
+            def __init__(
+                self,
+                model: torch.nn.Module,
+                score_threshold: float = 0.001,
+                iou_threshold: float = 0.7,
+                max_detections: int = 300,
+            ):
+                """
+                Wrapping PyTorch Module with multiclass_nms layer from sony_custom_layers.
+
+                Args:
+                    model (nn.Module): Model instance.
+                    score_threshold (float): Score threshold for non-maximum suppression.
+                    iou_threshold (float): Intersection over union threshold for non-maximum suppression.
+                    max_detections (float): The number of detections to return.
+                """
+                super().__init__()
+                self.model = model
+                self.score_threshold = score_threshold
+                self.iou_threshold = iou_threshold
+                self.max_detections = max_detections
+
+            def forward(self, images):
+                # model inference
+                outputs = self.model(images)
+
+                boxes = outputs[0]
+                scores = outputs[1]
+                nms = multiclass_nms(
+                    boxes=boxes,
+                    scores=scores,
+                    score_threshold=self.score_threshold,
+                    iou_threshold=self.iou_threshold,
+                    max_detections=self.max_detections,
+                )
+                return nms
+
+        quant_model = NMSWrapper(
+            model=quant_model,
+            score_threshold=self.args.conf or 0.001,
+            iou_threshold=self.args.iou,
+            max_detections=self.args.max_det,
+        ).to(self.device)
+
+        f = Path(str(self.file).replace(self.file.suffix, "_imx_model"))
+        f.mkdir(exist_ok=True)
+        onnx_model = f / Path(str(self.file).replace(self.file.suffix, "_imx.onnx"))  # js dir
+        mct.exporter.pytorch_export_model(
+            model=quant_model, save_model_path=onnx_model, repr_dataset=representative_dataset_gen
+        )
+
+        model_onnx = onnx.load(onnx_model)  # load onnx model
+        for k, v in self.metadata.items():
+            meta = model_onnx.metadata_props.add()
+            meta.key, meta.value = k, str(v)
+
+        onnx.save(model_onnx, onnx_model)
+
+        subprocess.run(
+            ["imxconv-pt", "-i", str(onnx_model), "-o", str(f), "--no-input-persistency", "--overwrite-output"],
+            check=True,
+        )
+
+        # Needed for imx models.
+        with open(f / "labels.txt", "w") as file:
+            file.writelines([f"{name}\n" for _, name in self.model.names.items()])
+
+        return f, None
+
     def _add_tflite_metadata(self, file):
         """Add metadata to *.tflite models per https://www.tensorflow.org/lite/models/convert/metadata."""
         import flatbuffers

ultralytics/engine/model.py

@@ -2,7 +2,7 @@
 
 import inspect
 from pathlib import Path
-from typing import List, Union
+from typing import Dict, List, Union
 
 import numpy as np
 import torch
@@ -881,7 +881,7 @@ class Model(nn.Module):
         return self
 
     @property
-    def names(self) -> list:
+    def names(self) -> Dict[int, str]:
         """
         Retrieves the class names associated with the loaded model.
 

ultralytics/engine/results.py

@@ -535,9 +535,9 @@ class Results(SimpleClass):
         # Plot Detect results
         if pred_boxes is not None and show_boxes:
             for i, d in enumerate(reversed(pred_boxes)):
-                c, conf, id = int(d.cls), float(d.conf) if conf else None, None if d.id is None else int(d.id.item())
+                c, d_conf, id = int(d.cls), float(d.conf) if conf else None, None if d.id is None else int(d.id.item())
                 name = ("" if id is None else f"id:{id} ") + names[c]
-                label = (f"{name} {conf:.2f}" if conf else name) if labels else None
+                label = (f"{name} {d_conf:.2f}" if conf else name) if labels else None
                 box = d.xyxyxyxy.reshape(-1, 4, 2).squeeze() if is_obb else d.xyxy.squeeze()
                 annotator.box_label(
                     box,

ultralytics/engine/trainer.py

@@ -792,7 +792,7 @@ class BaseTrainer:
                     g[0].append(param)
 
         optimizers = {"Adam", "Adamax", "AdamW", "NAdam", "RAdam", "RMSProp", "SGD", "auto"}
-        name = {x.lower(): x for x in optimizers}.get(name.lower(), None)
+        name = {x.lower(): x for x in optimizers}.get(name.lower())
         if name in {"Adam", "Adamax", "AdamW", "NAdam", "RAdam"}:
             optimizer = getattr(optim, name, optim.Adam)(g[2], lr=lr, betas=(momentum, 0.999), weight_decay=0.0)
         elif name == "RMSProp":

ultralytics/nn/autobackend.py

@@ -123,6 +123,7 @@ class AutoBackend(nn.Module):
             paddle,
             mnn,
             ncnn,
+            imx,
             triton,
         ) = self._model_type(w)
         fp16 &= pt or jit or onnx or xml or engine or nn_module or triton  # FP16
@@ -182,8 +183,8 @@ class AutoBackend(nn.Module):
             check_requirements("opencv-python>=4.5.4")
             net = cv2.dnn.readNetFromONNX(w)
 
-        # ONNX Runtime
-        elif onnx:
+        # ONNX Runtime and IMX
+        elif onnx or imx:
             LOGGER.info(f"Loading {w} for ONNX Runtime inference...")
             check_requirements(("onnx", "onnxruntime-gpu" if cuda else "onnxruntime"))
             if IS_RASPBERRYPI or IS_JETSON:
@@ -199,7 +200,22 @@ class AutoBackend(nn.Module):
                 device = torch.device("cpu")
                 cuda = False
             LOGGER.info(f"Preferring ONNX Runtime {providers[0]}")
-            session = onnxruntime.InferenceSession(w, providers=providers)
+            if onnx:
+                session = onnxruntime.InferenceSession(w, providers=providers)
+            else:
+                check_requirements(
+                    ["model-compression-toolkit==2.1.1", "sony-custom-layers[torch]==0.2.0", "onnxruntime-extensions"]
+                )
+                w = next(Path(w).glob("*.onnx"))
+                LOGGER.info(f"Loading {w} for ONNX IMX inference...")
+                import mct_quantizers as mctq
+                from sony_custom_layers.pytorch.object_detection import nms_ort  # noqa
+
+                session = onnxruntime.InferenceSession(
+                    w, mctq.get_ort_session_options(), providers=["CPUExecutionProvider"]
+                )
+                task = "detect"
+
             output_names = [x.name for x in session.get_outputs()]
             metadata = session.get_modelmeta().custom_metadata_map
             dynamic = isinstance(session.get_outputs()[0].shape[0], str)
@@ -520,7 +536,7 @@ class AutoBackend(nn.Module):
             y = self.net.forward()
 
         # ONNX Runtime
-        elif self.onnx:
+        elif self.onnx or self.imx:
             if self.dynamic:
                 im = im.cpu().numpy()  # torch to numpy
                 y = self.session.run(self.output_names, {self.session.get_inputs()[0].name: im})
@@ -537,6 +553,9 @@ class AutoBackend(nn.Module):
                 )
                 self.session.run_with_iobinding(self.io)
                 y = self.bindings
+            if self.imx:
+                # boxes, conf, cls
+                y = np.concatenate([y[0], y[1][:, :, None], y[2][:, :, None]], axis=-1)
 
         # OpenVINO
         elif self.xml:

ultralytics/nn/modules/block.py

@@ -240,7 +240,8 @@ class C2f(nn.Module):
 
     def forward_split(self, x):
         """Forward pass using split() instead of chunk()."""
-        y = list(self.cv1(x).split((self.c, self.c), 1))
+        y = self.cv1(x).split((self.c, self.c), 1)
+        y = [y[0], y[1]]
         y.extend(m(y[-1]) for m in self.m)
         return self.cv2(torch.cat(y, 1))
 

ultralytics/nn/modules/head.py

@@ -23,6 +23,7 @@ class Detect(nn.Module):
 
     dynamic = False  # force grid reconstruction
     export = False  # export mode
+    format = None  # export format
     end2end = False  # end2end
     max_det = 300  # max_det
     shape = None
@@ -101,7 +102,7 @@ class Detect(nn.Module):
         # Inference path
         shape = x[0].shape  # BCHW
         x_cat = torch.cat([xi.view(shape[0], self.no, -1) for xi in x], 2)
-        if self.dynamic or self.shape != shape:
+        if self.format != "imx" and (self.dynamic or self.shape != shape):
             self.anchors, self.strides = (x.transpose(0, 1) for x in make_anchors(x, self.stride, 0.5))
             self.shape = shape
 
@@ -119,6 +120,11 @@ class Detect(nn.Module):
             grid_size = torch.tensor([grid_w, grid_h, grid_w, grid_h], device=box.device).reshape(1, 4, 1)
             norm = self.strides / (self.stride[0] * grid_size)
             dbox = self.decode_bboxes(self.dfl(box) * norm, self.anchors.unsqueeze(0) * norm[:, :2])
+        elif self.export and self.format == "imx":
+            dbox = self.decode_bboxes(
+                self.dfl(box) * self.strides, self.anchors.unsqueeze(0) * self.strides, xywh=False
+            )
+            return dbox.transpose(1, 2), cls.sigmoid().permute(0, 2, 1)
         else:
             dbox = self.decode_bboxes(self.dfl(box), self.anchors.unsqueeze(0)) * self.strides
 
@@ -137,9 +143,9 @@ class Detect(nn.Module):
                 a[-1].bias.data[:] = 1.0  # box
                 b[-1].bias.data[: m.nc] = math.log(5 / m.nc / (640 / s) ** 2)  # cls (.01 objects, 80 classes, 640 img)
 
-    def decode_bboxes(self, bboxes, anchors):
+    def decode_bboxes(self, bboxes, anchors, xywh=True):
         """Decode bounding boxes."""
-        return dist2bbox(bboxes, anchors, xywh=not self.end2end, dim=1)
+        return dist2bbox(bboxes, anchors, xywh=xywh and (not self.end2end), dim=1)
 
     @staticmethod
     def postprocess(preds: torch.Tensor, max_det: int, nc: int = 80):

ultralytics/solutions/solutions.py

@@ -72,14 +72,13 @@ class BaseSolution:
         self.model = YOLO(self.CFG["model"])
         self.names = self.model.names
 
-        if IS_CLI:  # for CLI, download the source and init video writer
-            if self.CFG["source"] is None:
-                d_s = "solutions_ci_demo.mp4" if "-pose" not in self.CFG["model"] else "solution_ci_pose_demo.mp4"
-                LOGGER.warning(f"⚠️ WARNING: source not provided. using default source {ASSETS_URL}/{d_s}")
-                from ultralytics.utils.downloads import safe_download
-
-                safe_download(f"{ASSETS_URL}/{d_s}")  # download source from ultralytics assets
-                self.CFG["source"] = d_s  # set default source
+        if IS_CLI and self.CFG["source"] is None:
+            d_s = "solutions_ci_demo.mp4" if "-pose" not in self.CFG["model"] else "solution_ci_pose_demo.mp4"
+            LOGGER.warning(f"⚠️ WARNING: source not provided. using default source {ASSETS_URL}/{d_s}")
+            from ultralytics.utils.downloads import safe_download
+
+            safe_download(f"{ASSETS_URL}/{d_s}")  # download source from ultralytics assets
+            self.CFG["source"] = d_s  # set default source
 
         # Initialize environment and region setup
         self.env_check = check_imshow(warn=True)

ultralytics/utils/benchmarks.py

@@ -118,6 +118,11 @@ def benchmark(
                 assert not IS_JETSON, "MNN export not supported on NVIDIA Jetson"
             if i == 13:  # NCNN
                 assert not isinstance(model, YOLOWorld), "YOLOWorldv2 NCNN exports not supported yet"
+            if i == 14:  # 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"
             if "cpu" in device.type:
                 assert cpu, "inference not supported on CPU"
             if "cuda" in device.type:

ultralytics/utils/callbacks/comet.py

@@ -291,7 +291,7 @@ def _log_plots(experiment, trainer):
             for plots in EVALUATION_PLOT_NAMES
             for prefix in POSE_METRICS_PLOT_PREFIX
         ]
-    elif isinstance(trainer.validator.metrics, DetMetrics) or isinstance(trainer.validator.metrics, OBBMetrics):
+    elif isinstance(trainer.validator.metrics, (DetMetrics, OBBMetrics)):
         plot_filenames = [trainer.save_dir / f"{plots}.png" for plots in EVALUATION_PLOT_NAMES]
 
     if plot_filenames is not None:

ultralytics/utils/callbacks/raytune.py

@@ -16,8 +16,7 @@ def on_fit_epoch_end(trainer):
     """Sends training metrics to Ray Tune at end of each epoch."""
     if ray.train._internal.session._get_session():  # replacement for deprecated ray.tune.is_session_enabled()
         metrics = trainer.metrics
-        metrics["epoch"] = trainer.epoch
-        session.report(metrics)
+        session.report({**metrics, **{"epoch": trainer.epoch + 1}})
 
 
 callbacks = (

ultralytics/utils/tal.py

@@ -306,7 +306,7 @@ def make_anchors(feats, strides, grid_cell_offset=0.5):
     assert feats is not None
     dtype, device = feats[0].dtype, feats[0].device
     for i, stride in enumerate(strides):
-        _, _, h, w = feats[i].shape
+        h, w = feats[i].shape[2:] if isinstance(feats, list) else (int(feats[i][0]), int(feats[i][1]))
         sx = torch.arange(end=w, device=device, dtype=dtype) + grid_cell_offset  # shift x
         sy = torch.arange(end=h, device=device, dtype=dtype) + grid_cell_offset  # shift y
         sy, sx = torch.meshgrid(sy, sx, indexing="ij") if TORCH_1_10 else torch.meshgrid(sy, sx)

ultralytics/utils/torch_utils.py

@@ -729,3 +729,48 @@ class EarlyStopping:
                 f"i.e. `patience=300` or use `patience=0` to disable EarlyStopping."
             )
         return stop
+
+
+class FXModel(nn.Module):
+    """
+    A custom model class for torch.fx compatibility.
+
+    This class extends `torch.nn.Module` and is designed to ensure compatibility with torch.fx for tracing and graph manipulation.
+    It copies attributes from an existing model and explicitly sets the model attribute to ensure proper copying.
+
+    Args:
+        model (torch.nn.Module): The original model to wrap for torch.fx compatibility.
+    """
+
+    def __init__(self, model):
+        """
+        Initialize the FXModel.
+
+        Args:
+            model (torch.nn.Module): The original model to wrap for torch.fx compatibility.
+        """
+        super().__init__()
+        copy_attr(self, model)
+        # Explicitly set `model` since `copy_attr` somehow does not copy it.
+        self.model = model.model
+
+    def forward(self, x):
+        """
+        Forward pass through the model.
+
+        This method performs the forward pass through the model, handling the dependencies between layers and saving intermediate outputs.
+
+        Args:
+            x (torch.Tensor): The input tensor to the model.
+
+        Returns:
+            (torch.Tensor): The output tensor from the model.
+        """
+        y = []  # outputs
+        for m in self.model:
+            if m.f != -1:  # if not from previous layer
+                # from earlier layers
+                x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f]
+            x = m(x)  # run
+            y.append(x)  # save output
+        return x

{ultralytics-8.3.28.dist-info → ultralytics-8.3.29.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: ultralytics
-Version: 8.3.28
+Version: 8.3.29
 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>
@@ -96,7 +96,7 @@ Requires-Dist: streamlit; extra == "solutions"
 
 <div>
     <a href="https://github.com/ultralytics/ultralytics/actions/workflows/ci.yaml"><img src="https://github.com/ultralytics/ultralytics/actions/workflows/ci.yaml/badge.svg" alt="Ultralytics CI"></a>
-    <a href="https://pepy.tech/project/ultralytics"><img src="https://static.pepy.tech/badge/ultralytics" alt="Ultralytics Downloads"></a>
+    <a href="https://www.pepy.tech/projects/ultralytics"><img src="https://static.pepy.tech/badge/ultralytics" alt="Ultralytics Downloads"></a>
     <a href="https://zenodo.org/badge/latestdoi/264818686"><img src="https://zenodo.org/badge/264818686.svg" alt="Ultralytics YOLO Citation"></a>
     <a href="https://discord.com/invite/ultralytics"><img alt="Ultralytics Discord" src="https://img.shields.io/discord/1089800235347353640?logo=discord&logoColor=white&label=Discord&color=blue"></a>
     <a href="https://community.ultralytics.com/"><img alt="Ultralytics Forums" src="https://img.shields.io/discourse/users?server=https%3A%2F%2Fcommunity.ultralytics.com&logo=discourse&label=Forums&color=blue"></a>
@@ -143,7 +143,7 @@ See below for a quickstart install and usage examples, and see our [Docs](https:
 
 Pip install the ultralytics package including all [requirements](https://github.com/ultralytics/ultralytics/blob/main/pyproject.toml) in a [**Python>=3.8**](https://www.python.org/) environment with [**PyTorch>=1.8**](https://pytorch.org/get-started/locally/).
 
-[![PyPI - Version](https://img.shields.io/pypi/v/ultralytics?logo=pypi&logoColor=white)](https://pypi.org/project/ultralytics/) [![Ultralytics Downloads](https://static.pepy.tech/badge/ultralytics)](https://pepy.tech/project/ultralytics) [![PyPI - Python Version](https://img.shields.io/pypi/pyversions/ultralytics?logo=python&logoColor=gold)](https://pypi.org/project/ultralytics/)
+[![PyPI - Version](https://img.shields.io/pypi/v/ultralytics?logo=pypi&logoColor=white)](https://pypi.org/project/ultralytics/) [![Ultralytics Downloads](https://static.pepy.tech/badge/ultralytics)](https://www.pepy.tech/projects/ultralytics) [![PyPI - Python Version](https://img.shields.io/pypi/pyversions/ultralytics?logo=python&logoColor=gold)](https://pypi.org/project/ultralytics/)
 
 ```bash
 pip install ultralytics

{ultralytics-8.3.28.dist-info → ultralytics-8.3.29.dist-info}/RECORD

@@ -3,11 +3,11 @@ tests/conftest.py,sha256=9PFAiwAy6eeORGspr5dOKxVuFDVKqYg8Nn_RxSJ27UI,2919
 tests/test_cli.py,sha256=G7OJ1ErQYsGy2Dx1zP-0p7EZR4aPoAdtLGiY4Hm7jQM,5006
 tests/test_cuda.py,sha256=rhHFvKNegN1ChtueKM0JhATJaJDFB377uXo2Kca5JVQ,5943
 tests/test_engine.py,sha256=dcEcJsMQh61rDSNv7l4TIAgybLpzjVwerv9JZC_KCM8,4934
-tests/test_exports.py,sha256=lE5P5Fftd7z-tThSNJHNI5UTchg_RntxFkxrnhmUHZM,8389
+tests/test_exports.py,sha256=1MvhcQ2qHdbJImHII-bFarcaIcm-kPlEK-OdFLxnj7o,8769
 tests/test_integrations.py,sha256=f5-QCUk1SU_-qn4mBCZwS3GN3tXEBIIXo4z2EhExbHw,6126
 tests/test_python.py,sha256=I1RRdCwLdrc3jX06huVxct8HX8ccQOmQgVpuEflRl0U,23560
 tests/test_solutions.py,sha256=sPYhy2d814mIVvojQeVxeZPu0IVy01_Y8zuMcu_9GF0,3790
-ultralytics/__init__.py,sha256=xtS8JoiE1smNjmmioTji7vWUpxcUOkx84jZEIAlxISs,681
+ultralytics/__init__.py,sha256=DQQhUIqALl4beZ1ywcZyZ7EXgENfKTLSTzAOl1ctiOY,681
 ultralytics/assets/bus.jpg,sha256=wCAZxJecGR63Od3ZRERe9Aja1Weayrb9Ug751DS_vGM,137419
 ultralytics/assets/zidane.jpg,sha256=Ftc4aeMmen1O0A3o6GCDO9FlfBslLpTAw0gnetx7bts,50427
 ultralytics/cfg/__init__.py,sha256=0X6rETee3FHzNENaPrkByFi7dtpj91x4PCYF1-RxKdI,38633
@@ -93,17 +93,17 @@ ultralytics/data/annotator.py,sha256=JNmS6uELlEABrU5ViVJiPnjt44v-Us7j39Bwoug_73Y
 ultralytics/data/augment.py,sha256=YCLrwx1mRGeidggo_7GeINay8KdxACqREHJofZeaTHA,120430
 ultralytics/data/base.py,sha256=ZCIhAyFfxXVp5fVnYD8mwbksNALJTayBKIR5FKGV7ZM,15168
 ultralytics/data/build.py,sha256=AfMmz0sHIYmwry_90tEJFRk_kz0S3SolScVXqYHiT08,7261
-ultralytics/data/converter.py,sha256=wd1u2BPZQRAkobv8ThklpeQAfcYL2xo-i94qNJdbwBU,24286
+ultralytics/data/converter.py,sha256=RIfTXNrazwZqmTYOYoJtupDMtNzm8dxsrVp6q2m8gyg,24388
 ultralytics/data/dataset.py,sha256=D556AW0ZEsW3V8c5zJiHM_prc_YfZqymIkDKPw3k9Io,22936
 ultralytics/data/loaders.py,sha256=Fr70Q9p9t7buLW_8R2_lI_nyCMG033gWSxvwy1M-a-U,28449
 ultralytics/data/split_dota.py,sha256=eFafJ7Vg52wj6KDCHFJAf1tKzyPD5YaPB8kM4VX5Aeg,10688
 ultralytics/data/utils.py,sha256=bmWEIrdogj4kssZQSJdSbIF8QsJU00lo-EY-Mgcqv4M,31073
 ultralytics/engine/__init__.py,sha256=mHtJuK4hwF8cuV-VHDc7tp6u6D1gHz2Z7JI8grmQDTs,42
-ultralytics/engine/exporter.py,sha256=Bcv_TiMYN6f7dYeLjhOQZjuqXqASYUX0c3vy93sUTuI,60519
-ultralytics/engine/model.py,sha256=pvL1uf-wwdWL8Iph7VEAYn1-z7wEHzVug21V_0_gO6M,51456
+ultralytics/engine/exporter.py,sha256=DH67LwNDr3fiWxaES-lhSLvm5pCuasXLbQv4FSLCi_M,67171
+ultralytics/engine/model.py,sha256=TfuTczFjNJ3GW0E_qWVH6OaJ_2I-_Srx7i_4GQebDoo,51472
 ultralytics/engine/predictor.py,sha256=aS4yJdTK2kYq-TTpzIlWxqnAcBz38zIECZoMb_yOPMY,17597
-ultralytics/engine/results.py,sha256=BxanBI8PhBCfs-9cSy-GS6naScuiD3hdvUAJWPW2mS0,75043
-ultralytics/engine/trainer.py,sha256=eyIKlUdPuvKKWpqsUrRqP7mfj1CAHIPzf5MYjYmqwGA,37155
+ultralytics/engine/results.py,sha256=BZVQF8TbNRnf2DcnTYzVCin1NlpplWaEW9EskACvhOI,75047
+ultralytics/engine/trainer.py,sha256=lbFMLdrdWkk1td6BpUS0_uLhAkiWo-eAmx_Kaov1JPA,37149
 ultralytics/engine/tuner.py,sha256=WBj8iw1K1TK0hvanlA-wkwmfqh1SI8jEe2dGwUINeTg,11838
 ultralytics/engine/validator.py,sha256=aWpXE3nrOqaA7jCuUgwxi0FabiGTIXtZvjoJyCX903o,14870
 ultralytics/hub/__init__.py,sha256=c6Me4E8V-P7mtzTggyPYz9FnVkqWRyPp9F-fMcyFNQ0,5632
@@ -169,13 +169,13 @@ ultralytics/models/yolo/world/__init__.py,sha256=3VTH0q4NOt2EWRom15yCymvmvm0Etp2
 ultralytics/models/yolo/world/train.py,sha256=gaDrAmLJpg9qDtmL5evA5HsV2yb4RTRSfk2EDYrHdRg,3686
 ultralytics/models/yolo/world/train_world.py,sha256=IsnCEVt6DcM9lUskCKmIN-M8MM79xLpwTRqRoAHUnZ4,4857
 ultralytics/nn/__init__.py,sha256=4BPLHY89xEM_al5uK0aOmFgiML6CMGEZbezxOvTjOEs,587
-ultralytics/nn/autobackend.py,sha256=TZdpKEtSAINAkXzNF_b5rG4c_mjnkUtNnQ2Ux1reSqM,34766
+ultralytics/nn/autobackend.py,sha256=Arke5BaRQmr4yQd-xr6Z8P7kbTBNLI-O0fsDPFLOXMw,35625
 ultralytics/nn/tasks.py,sha256=NWe0cL7A0LpsP3S1Efvi2NutAjWc_rGYMJMwAeb2bAg,48605
 ultralytics/nn/modules/__init__.py,sha256=xhW2BennT9U_VaMXVpRu-bdLgp1BXt9L8mkIUBE3idU,2625
 ultralytics/nn/modules/activation.py,sha256=chhn469wnRHEs5BMGNBYXwPYZc_7-urspTT8fnBd-xA,895
-ultralytics/nn/modules/block.py,sha256=thcIPcnGRRxDDDswywJsfzbewr9XfTrzl_UvSl-bJ3c,41832
+ultralytics/nn/modules/block.py,sha256=PAm23KpRHDNlGtNWf1w8Ae0LdjII2H5vu0A4eeWx_XQ,41851
 ultralytics/nn/modules/conv.py,sha256=vOeHZ6Z4sc6-9PrDmRGT1hFkxSBbbWkQm2jRbGGjpqQ,12705
-ultralytics/nn/modules/head.py,sha256=3ULpEpr2_I4bd9JSptX_9zRKimdTOm4y8qT-DG-Gzq4,27456
+ultralytics/nn/modules/head.py,sha256=KCO-qarg2K7uJqQ7L5zVJ4-viiHqmu4bzbSgAw3L_nk,27815
 ultralytics/nn/modules/transformer.py,sha256=tGiK8NmPfswwW1rbF21r5ILUkkZQ6Nk4s8j16vFBmps,18069
 ultralytics/nn/modules/utils.py,sha256=a88cKl2wz1nMVSEBiajtvaCbDBQIkESWOKTZ_WAJy90,3195
 ultralytics/solutions/__init__.py,sha256=6RDeXWO1QSaMgCq8YrWXaj2xvPw2sJwJL_a0dgjCvz0,648
@@ -186,7 +186,7 @@ ultralytics/solutions/heatmap.py,sha256=If9rosSCmE7pAL1HtVnLkx05gQp6nP1K6HzATMca
 ultralytics/solutions/object_counter.py,sha256=vKB7riRm8NjHA6IXyf557FpmV-b0_XoKbXHqMHziXSM,8264
 ultralytics/solutions/parking_management.py,sha256=1DsEE94eauqcnnFxUYI-BX9eA1GbJVNt7oncj1okYpI,11198
 ultralytics/solutions/queue_management.py,sha256=D9TqwJSVrZQFxp_M8O62WfBAxkAuDWWnXe7FFmnp7_w,4881
-ultralytics/solutions/solutions.py,sha256=wB-w0URApoaykrg0a2XzCCEzODMkeBqbG2xA3iWmeLg,7294
+ultralytics/solutions/solutions.py,sha256=q2nR5J9vJTQfuMHEuxdor1MhbQTP1WoCh9GmoXiKxcY,7208
 ultralytics/solutions/speed_estimation.py,sha256=A10DmuZlGkoZUyfHhZWcDRjj1-9GXiDhEjyBbAzfaDs,4936
 ultralytics/solutions/streamlit_inference.py,sha256=w4dnvSv2FOrpji9W1Ir86phka3OXc7jd_38-OCbQdZw,5701
 ultralytics/trackers/__init__.py,sha256=j72IgH2dZHQArMPK4YwcV5ieIw94fYvlGdQjB9cOQKw,227
@@ -200,7 +200,7 @@ ultralytics/trackers/utils/kalman_filter.py,sha256=cH9zD3fwkuezP97H9mw8cSBN7a8hH
 ultralytics/trackers/utils/matching.py,sha256=3Ie1WNNRZ4_q3365F03XD7Nr9juZB_08mw4yUKC3w74,7162
 ultralytics/utils/__init__.py,sha256=08pFkzKn1eR9xdIFhx8tx_8MO-gqXjt2n0HGwDeUlWE,49159
 ultralytics/utils/autobatch.py,sha256=BO9MCRtrLDtrDQaxqV0BdjaYsgXf-q07Y3_VdGp4URY,4330
-ultralytics/utils/benchmarks.py,sha256=tKpLuxHYJDmhN98E8jUsX7xWtFn1w1LuoQjwLlM76tA,25459
+ultralytics/utils/benchmarks.py,sha256=aEW28iVIMj-8bwOgISDphOJExDmaGi5bz3G2PJlRjcc,25793
 ultralytics/utils/checks.py,sha256=KXQSeauhzecy9tSjyDVy8oXbTDkHSSB9lOTYrqRWpok,29582
 ultralytics/utils/dist.py,sha256=NDFga-uKxkBX2zLxFHSene_cCiGQJoyOeCXcN9JIOIk,2358
 ultralytics/utils/downloads.py,sha256=fh7I5toTSowAOXtmx5zIzCEDREfTFG45cLIHmsDmuYw,21974
@@ -212,24 +212,24 @@ ultralytics/utils/metrics.py,sha256=msPaXc244ndc0NPBhnNlHsKkVhdc-TMgFn5NATlZZVI,
 ultralytics/utils/ops.py,sha256=dsXNdyrYx_p6io6zezig9p84dxS7U-10vceHNVu2IL0,32888
 ultralytics/utils/patches.py,sha256=J-iOwIRbfUs-inBZerhnXby5tUKjYcOIyvhLTS352JE,3270
 ultralytics/utils/plotting.py,sha256=TKtdbAOl6gZdFD2hlA5T4LNWfr2LUWbCC-cXkgL1JAU,61089
-ultralytics/utils/tal.py,sha256=ECsu95xEqOItmxMDN4YTD3FsUiIsQNWy0pZC3TfvFfk,16877
-ultralytics/utils/torch_utils.py,sha256=91fmJtZRvIVb6LI-wNkNrlHE7mMNBmcR4oif8ZYppYU,30089
+ultralytics/utils/tal.py,sha256=89m5adNGmwwFlUx895b_7lEjIJc8YBdivJaxl6ACaSA,16944
+ultralytics/utils/torch_utils.py,sha256=jB03Q-9ajTplxE05CdkmJmpXDUkb4LSiv3S6S2laWII,31608
 ultralytics/utils/triton.py,sha256=gg1finxno_tY2Ge9PMhmu7PI9wvoFZoiicdT4Bhqv3w,3936
 ultralytics/utils/tuner.py,sha256=K09-z5k1E4ZriSKoWdwQrJ2PJ2fY1ez3-b2R6aKPTqM,6198
 ultralytics/utils/callbacks/__init__.py,sha256=YrWqC3BVVaTLob4iCPR6I36mUxIUOpPJW7B_LjT78Qw,214
 ultralytics/utils/callbacks/base.py,sha256=PHjQ6RITwC2dylCQTB0bdPgAsHjxVeuDb5N1NPTbHGc,5775
 ultralytics/utils/callbacks/clearml.py,sha256=qbLbqzMVWAnjqg5YUM-Ue6CmGueFCvqKpHFKlw-MyVc,5933
-ultralytics/utils/callbacks/comet.py,sha256=XGMgBAfqqlGeJSF4OO_QUSaydNMY8BgLytBVeeISCl8,15057
+ultralytics/utils/callbacks/comet.py,sha256=EzSraWdMf54HPtt0xprHfudhITBkMTZHlT7wObCIA9c,15018
 ultralytics/utils/callbacks/dvc.py,sha256=WIClMsuvhiiyrwRv5BsZLxjsxYNJ3Y8Vq7zN0Bthtro,5045
 ultralytics/utils/callbacks/hub.py,sha256=EPewsLigFQc9ucTX2exKSlKBiaBNhYYyGC_nR2ragJo,3997
 ultralytics/utils/callbacks/mlflow.py,sha256=mkl_rK0Gy02cXnQUYmzmLE5W97fMgfEb7IlgOAdnjHg,5396
 ultralytics/utils/callbacks/neptune.py,sha256=IbGQfEltamUKXJt93uSLQFn8c2rYh3DMTgVE1xsnmUI,3813
-ultralytics/utils/callbacks/raytune.py,sha256=ODVYzy-CoM4Uge0zjkh3Hnh9nF2M0vhDrSenXnvcizw,705
+ultralytics/utils/callbacks/raytune.py,sha256=Ck_yFzg7UZXiDWrLHaltjQybzVWSFDfzpdrx9ZYTRfI,700
 ultralytics/utils/callbacks/tensorboard.py,sha256=SHlE58Fb-sg-uZKtgy-ybIO3SAIfK55aj8kTYGA0Cyg,4167
 ultralytics/utils/callbacks/wb.py,sha256=oX3JarCJGhzvW556XiEXQNaZblAaK_UETAt3kzkY61w,6869
-ultralytics-8.3.28.dist-info/LICENSE,sha256=DZak_2itbUtvHzD3E7GNUYSRK6jdOJ-GqncQ2weavLA,34523
-ultralytics-8.3.28.dist-info/METADATA,sha256=IbIIi30q4VHA1_C5zJXiQAQ5HQQ1yFNl0nWWaH6_qUQ,35203
-ultralytics-8.3.28.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
-ultralytics-8.3.28.dist-info/entry_points.txt,sha256=YM_wiKyTe9yRrsEfqvYolNO5ngwfoL4-NwgKzc8_7sI,93
-ultralytics-8.3.28.dist-info/top_level.txt,sha256=XP49TwiMw4QGsvTLSYiJhz1xF_k7ev5mQ8jJXaXi45Q,12
-ultralytics-8.3.28.dist-info/RECORD,,
+ultralytics-8.3.29.dist-info/LICENSE,sha256=DZak_2itbUtvHzD3E7GNUYSRK6jdOJ-GqncQ2weavLA,34523
+ultralytics-8.3.29.dist-info/METADATA,sha256=yh-DydnZ0WaLyeF3GoeuT1Z8NJsVBOV3iy00xBLDmTs,35213
+ultralytics-8.3.29.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
+ultralytics-8.3.29.dist-info/entry_points.txt,sha256=YM_wiKyTe9yRrsEfqvYolNO5ngwfoL4-NwgKzc8_7sI,93
+ultralytics-8.3.29.dist-info/top_level.txt,sha256=XP49TwiMw4QGsvTLSYiJhz1xF_k7ev5mQ8jJXaXi45Q,12
+ultralytics-8.3.29.dist-info/RECORD,,