[YOLOV8] Update version

setup.py

@@ -133,7 +133,7 @@ def _parse_requirements_file(file_path):
     "opencv-python<=4.6.0.66",
 ]
 _yolo_integration_deps = [
-    "torchvision>=0.3.0,<0.14",
+    "torchvision>=0.3.0,<=0.15.1",
     "opencv-python<=4.6.0.66",
 ]
 _openpifpaf_integration_deps = [
@@ -142,7 +142,7 @@ def _parse_requirements_file(file_path):
     "pycocotools >=2.0.6",
     "scipy==1.10.1",
 ]
-_yolov8_integration_deps = _yolo_integration_deps + ["ultralytics==8.0.30"]
+_yolov8_integration_deps = _yolo_integration_deps + ["ultralytics==8.0.124"]
 _transformers_integration_deps = [
     f"{'nm-transformers' if is_release else 'nm-transformers-nightly'}"
     f"~={version_base}",

src/deepsparse/yolov8/pipelines.py

@@ -108,20 +108,14 @@ def process_engine_outputs(
             )
 
     def process_engine_outputs_seg(
-        self, engine_outputs: List[numpy.ndarray], nm: int = 32, **kwargs
+        self, engine_outputs: List[numpy.ndarray], nc: int = 80, **kwargs
     ) -> YOLOSegOutput:
         """
         The pathway for processing the outputs of the engine for YOLOv8 segmentation.
         :param engine_outputs: list of numpy arrays that are the output of the engine
             forward pass
-        :param nm: number of mask protos. It is information needed to perform NMS on the
-            segmentation output. It can be calculated in the following way:
-                dims = bboxes + classes + nm
-                where:
-                - bboxes = 4
-                - classes = 80 (explicit assumption -> we are using COCO dataset)
-                - dims = 116 (comes from the fact that
-                    detection.shape == [B, dims, 8400])
+        : params nc: number of classes. If not provided, calculated as
+            detection.shape[1] - 4
         :return: outputs of engine post-processed into an object in the `output_schema`
             format of this pipeline
         """
@@ -131,9 +125,9 @@ def process_engine_outputs_seg(
         # NMS
         detections_output = self.nms_function(
             outputs=detections,
+            nc=nc,
             iou_thres=kwargs.get("iou_thres", 0.25),
             conf_thres=kwargs.get("conf_thres", 0.45),
-            nm=nm,
             multi_label=kwargs.get("multi_label", False),
         )
 

src/deepsparse/yolov8/utils/validation/deepsparse_validator.py

@@ -9,14 +9,7 @@
 
 from deepsparse.yolov8.utils.validation.helpers import schema_to_tensor
 from ultralytics.yolo.data.utils import check_det_dataset
-from ultralytics.yolo.utils import (
-    DEFAULT_CFG,
-    LOGGER,
-    RANK,
-    SETTINGS,
-    TQDM_BAR_FORMAT,
-    callbacks,
-)
+from ultralytics.yolo.utils import LOGGER, TQDM_BAR_FORMAT, callbacks
 from ultralytics.yolo.utils.ops import Profile
 from ultralytics.yolo.utils.torch_utils import select_device, smart_inference_mode
 
@@ -42,7 +35,6 @@ def __call__(self, classes: Dict[int, str]):
         callbacks.add_integration_callbacks(self)
         self.run_callbacks("on_val_start")
         self.device = select_device(self.args.device, self.args.batch)
-        self.args.half &= self.device.type != "cpu"
         self.data = check_det_dataset(self.args.data)
         if isinstance(self.data["path"], str):
             self.data["path"] = Path(self.data["path"])
@@ -53,7 +45,7 @@ def __call__(self, classes: Dict[int, str]):
             )
 
         self.dataloader = self.dataloader or self.get_dataloader(
-            self.data.get("val") or self.data.set("test"), self.args.batch
+            self.data.get(self.args.split), self.args.batch
         )
 
         dt = Profile(), Profile(), Profile(), Profile()
@@ -93,18 +85,22 @@ def __call__(self, classes: Dict[int, str]):
         stats = self.get_stats()
         self.check_stats(stats)
         self.print_results()
-        self.speed = tuple(
-            x.t / len(self.dataloader.dataset) * 1e3 for x in dt
+        self.speed = dict(
+            zip(
+                self.speed.keys(),
+                (x.t / len(self.dataloader.dataset) * 1e3 for x in dt),
+            )
         )  # speeds per image
+        self.finalize_metrics()
         self.run_callbacks("on_val_end")
 
-        self.logger.info(
+        LOGGER.info(
             "Speed: %.1fms pre-process, %.1fms inference, %.1fms loss, %.1fms post-process per image"
-            % self.speed
+            % tuple(self.speed.values())
         )
         if self.args.save_json and self.jdict:
             with open(str(self.save_dir / "predictions.json"), "w") as f:
-                self.logger.info(f"Saving {f.name}...")
+                LOGGER.info(f"Saving {f.name}...")
                 json.dump(self.jdict, f)  # flatten and save
             stats = self.eval_json(stats)  # update stats
         return stats

src/deepsparse/yolov8/utils/validation/detection_validator.py

@@ -23,10 +23,9 @@ def __init__(
         dataloader=None,
         save_dir=None,
         pbar=None,
-        logger=None,
         args=None,
     ):
-        DetectionValidator.__init__(self, dataloader, save_dir, pbar, logger, args)
+        DetectionValidator.__init__(self, dataloader, save_dir, pbar, args)
         DeepSparseValidator.__init__(self, pipeline)
 
     # deepsparse edit: replaced argument `model` with `classes`

src/deepsparse/yolov8/utils/validation/segmentation_validator.py

@@ -27,7 +27,7 @@ def __init__(
         logger=None,
         args=None,
     ):
-        SegmentationValidator.__init__(self, dataloader, save_dir, pbar, logger, args)
+        SegmentationValidator.__init__(self, dataloader, save_dir, pbar, args)
         DeepSparseValidator.__init__(self, pipeline)
 
     # deepsparse edit: replaced argument `model` with `classes`

src/deepsparse/yolov8/validation.py

@@ -84,7 +84,7 @@
 @click.option(
     "--subtask",
     default="detection",
-    type=click.Choice(["detection", "classification", "segmentation"]),
+    type=click.Choice(["detection", "segmentation"]),
     show_default=True,
     help="A subtask of YOLOv8 to run. Default is `detection`.",
 )