Add Tensorboard as dependency

Author: gdlg

lib/pyproject.toml

@@ -51,6 +51,7 @@ dependencies = [
     "nncf==2.17.0",
     "anomalib[core]==1.1.3",
     "numpy<2.0.0",
+    "tensorboardX>=1.8",
 ]
 
 [project.optional-dependencies]

lib/src/otx/data/dataset/detection_new.py

@@ -25,8 +25,9 @@ def __init__(self, dm_subset: Dataset, **kwargs) -> None:
         Args:
             **kwargs: Keyword arguments to pass to OTXDataset
         """
-        kwargs["sample_type"] = DetectionSample
-        super().__init__(dm_subset, **kwargs)
+        sample_type = DetectionSample
+        dm_subset = dm_subset.convert_to_schema(sample_type)
+        super().__init__(dm_subset=dm_subset, sample_type=sample_type, **kwargs)
 
         labels = dm_subset.schema.attributes["label"].categories.labels
         self.label_info = LabelInfo(

lib/src/otx/data/dataset/segmentation_new.py

@@ -20,8 +20,8 @@ class OTXSegmentationDataset(OTXDataset):
 
     def __init__(self, dm_subset: Dataset, **kwargs) -> None:
         sample_type = SegmentationSample
-        super().__init__(dm_subset=dm_subset, sample_type=sample_type, **kwargs)
         dm_subset = dm_subset.convert_to_schema(sample_type)
+        super().__init__(dm_subset=dm_subset, sample_type=sample_type, **kwargs)
 
         labels = dm_subset.schema.attributes["masks"].categories.labels
         self.label_info = SegLabelInfo(

lib/src/otx/data/entity/sample.py

@@ -131,23 +131,8 @@ class DetectionSample(OTXSample):
     """DetectionSample is a base class for OTX detection items."""
 
     image: np.ndarray | tv_tensors.Image = image_field(dtype=pl.UInt8)
-    label: torch.Tensor = label_field(pl.Int32(), is_list=True)
-    bboxes: torch.Tensor = bbox_field(dtype=pl.Float32)
-
-    @classmethod
-    def from_dm_item(cls, item: DatasetItem) -> DetectionSample:
-        """Create a DetectionSample from a Datumaro DatasetItem.
-
-        Args:
-            item: Datumaro DatasetItem containing image and label
-
-        Returns:
-            DetectionSample: Instance with image and label set
-        """
-        image = item.media_as(Image).data
-        bboxes = [bbox.points for bbox in item.annotations] if item.annotations else None
-        labels = [bbox.label for bbox in item.annotations] if item.annotations else None
-        return cls(image=image, label=labels, bboxes=bboxes)
+    label: np.ndarray | torch.Tensor = label_field(pl.Int32(), is_list=True)
+    bboxes: np.ndarray | tv_tensors.BoundingBoxes = bbox_field(dtype=pl.Float32)
 
     def __post_init__(self) -> None:
         shape = self.image.shape[:2]

lib/src/otx/data/factory.py

@@ -103,18 +103,11 @@ def create(
             return OTXHlabelClsDataset(**common_kwargs)
 
         if task == OTXTaskType.DETECTION:
-            from .dataset.detection_new import DetectionSample, OTXDetectionDataset
+            from .dataset.detection_new import OTXDetectionDataset
+
+            dataset = convert_from_legacy(dm_subset)
+            common_kwargs["dm_subset"] = dataset
 
-            if isinstance(dm_subset, DmDataset):
-                categories = cls._get_label_categories(dm_subset, data_format)
-                dataset = DatasetNew(DetectionSample, categories={"label": categories})
-                for item in dm_subset:
-                    if len(item.media.data.shape) == 3:
-                        dataset.append(DetectionSample.from_dm_item(item))
-                common_kwargs["dm_subset"] = dataset
-            else:
-                msg = "Dataset must be of type DmDataset."
-                raise RuntimeError(msg)
             return OTXDetectionDataset(**common_kwargs)
 
         if task in [OTXTaskType.ROTATED_DETECTION, OTXTaskType.INSTANCE_SEGMENTATION]: