Implement an AutoDataset (DAT-911)

datamint/dataset/__init__.py

@@ -6,7 +6,7 @@
 - VideoDataset: Temporal sequences (videos, multi-frame DICOM)
 - VolumetricDataset: 3D volumes (NIfTI, CT, MRI)
 
-Use `create_dataset()` for automatic type detection, or instantiate directly.
+Use `build_dataset()` for automatic type detection, or instantiate directly.
 """
 
 # New modular architecture
@@ -18,6 +18,7 @@
 from .sliced_dataset import SlicedVolumeDataset
 from .sliced_video_dataset import SlicedVideoDataset
 from .detection_dataset import DetectionDataset, detection_collate_fn
+from .factory import build_dataset
 
 __all__ = [
     # Core
@@ -32,4 +33,6 @@
     'SlicedVideoDataset',
     'DetectionDataset',
     'detection_collate_fn',
+    # Factory
+    'build_dataset',
 ]

datamint/dataset/factory.py

@@ -0,0 +1,99 @@
+from __future__ import annotations
+
+import logging
+from typing import TYPE_CHECKING, Any
+
+if TYPE_CHECKING:
+    from datamint.entities.resource import Resource
+    from .base import DatamintBaseDataset
+
+_LOGGER = logging.getLogger(__name__)
+
+
+def _classify_resource(resource: 'Resource') -> str:
+    if resource.is_video():
+        return 'video'
+    if resource.is_volume():
+        try:
+            if resource.get_depth() == 1:
+                return 'image'
+        except Exception:
+            pass  # frame_count not in list-response metadata; assume volume
+        return 'volume'
+    if resource.is_image():
+        return 'image'
+    return getattr(resource, 'kind', 'unknown')
+
+
+def build_dataset(project_name: str, **kwargs: Any) -> 'DatamintBaseDataset':
+    """Auto-detect and return the appropriate dataset class for a project.
+
+    Fetches a small sample of resources from the project to determine the
+    data type, then instantiates and returns the matching dataset class:
+
+    - 2D images (JPEG, PNG) → :class:`~datamint.dataset.ImageDataset`
+    - Volumes (NIfTI, DICOM, volumetric) → :class:`~datamint.dataset.VolumeDataset`
+    - Videos → :class:`~datamint.dataset.VideoDataset`
+
+    Args:
+        project_name: Name of the Datamint project.
+        **kwargs: Forwarded to the dataset constructor (transforms, filters, etc.).
+
+    Returns:
+        An instantiated dataset of the detected type.
+
+    Raises:
+        ValueError: If the project is empty, contains unknown resource types,
+            or contains a mix of data types.
+
+    Example::
+
+        from datamint.dataset import build_dataset
+
+        ds = build_dataset('MyProject', include_unannotated=False)
+    """
+    from datamint import Api
+    from .image_dataset import ImageDataset
+    from .volume_dataset import VolumeDataset
+    from .video_dataset import VideoDataset
+
+    _KIND_TO_CLS = {
+        'image': ImageDataset,
+        'volume': VolumeDataset,
+        'nifti': VolumeDataset,
+        'dicom': VolumeDataset,
+        'video': VideoDataset,
+    }
+
+    api = Api()
+    sample = api.resources.get_list(project_name=project_name, limit=5)
+
+    if not sample:
+        raise ValueError(f"Project '{project_name}' has no resources.")
+
+    kinds = {_classify_resource(r) for r in sample}
+    unknown = kinds - set(_KIND_TO_CLS)
+    if unknown:
+        raise ValueError(
+            f"Project '{project_name}' contains unsupported resource types: {sorted(unknown)}. "
+            "Instantiate the dataset class directly."
+        )
+
+    if kinds == {'image', 'volume'}:
+        _LOGGER.warning(
+            f"Project '{project_name}' contains a mix of 2D and 3D DICOM resources. "
+            "Defaulting to VolumeDataset. Use ImageDataset or VolumeDataset directly "
+            "if you need a specific type."
+        )
+        kinds = {'volume'}
+
+    if len(kinds) > 1:
+        raise ValueError(
+            f"Project '{project_name}' contains mixed data types: {sorted(kinds)}. "
+            "Instantiate the dataset class directly."
+        )
+
+    kind = next(iter(kinds))
+    dataset_cls = _KIND_TO_CLS[kind]
+    _LOGGER.info(f"Detected resource type '{kind}'; using {dataset_cls.__name__}.")
+    return dataset_cls(project=project_name, **kwargs)