Stop passing classes and default_classes to export and evaluation methods

docs/source/integrations/coco.rst

@@ -325,7 +325,8 @@ to add them to your dataset as follows:
     ]
 
     # Add COCO predictions to `predictions` field of dataset
-    fouc.add_coco_labels(coco_dataset, "predictions", predictions)
+    classes = coco_dataset.default_classes
+    fouc.add_coco_labels(coco_dataset, "predictions", predictions, classes)
 
     # Verify that predictions were added to two images
     print(coco_dataset.count("predictions"))  # 2

docs/source/integrations/cvat.rst

@@ -431,9 +431,8 @@ details:
     `label_field` or all fields in `label_schema` without classes specified.
     All new label fields must have a class list provided via one of the
     supported methods. For existing label fields, if classes are not provided
-    by this argument nor `label_schema`, they are retrieved from
-    :meth:`Dataset.get_classes() <fiftyone.core.dataset.Dataset.get_classes>`
-    if possible, or else the observed labels on your dataset are used
+    by this argument nor `label_schema`, the observed labels on your dataset
+    are used
 -   **attributes** (*True*): specifies the label attributes of each label field
     to include (other than their `label`, which is always included) in the
     annotation export. Can be any of the following:
@@ -631,16 +630,8 @@ FiftyOne can infer the appropriate values to use:
 
 -   **label_type**: if omitted, the |Label| type of the field will be used to
     infer the appropriate value for this parameter
--   **classes**: if omitted for a non-semantic segmentation field, the class
-    lists from the :meth:`classes <fiftyone.core.dataset.Dataset.classes>` or
-    :meth:`default_classes <fiftyone.core.dataset.Dataset.default_classes>`
-    properties of your dataset will be used, if available. Otherwise, the
-    observed labels on your dataset will be used to construct a classes list
--   **mask_targets**: if omitted for a semantic segmentation field, the mask
-    targets from the
-    :meth:`mask_targets <fiftyone.core.dataset.Dataset.mask_targets>` or
-    :meth:`default_mask_targets <fiftyone.core.dataset.Dataset.default_mask_targets>`
-    properties of your dataset will be used, if available
+-   **classes**: if omitted for a non-semantic segmentation field, the observed
+    labels on your dataset will be used to construct a classes list
 
 .. _cvat-label-attributes:
 

docs/source/user_guide/annotation.rst

@@ -472,9 +472,8 @@ more details:
     `label_field` or all fields in `label_schema` without classes specified.
     All new label fields must have a class list provided via one of the
     supported methods. For existing label fields, if classes are not provided
-    by this argument nor `label_schema`, they are retrieved from
-    :meth:`Dataset.get_classes() <fiftyone.core.dataset.Dataset.get_classes>`
-    if possible, or else the observed labels on your dataset are used
+    by this argument nor `label_schema`, the observed labels on your dataset
+    are used
 -   **attributes** (*True*): specifies the label attributes of each label field
     to include (other than their `label`, which is always included) in the
     annotation export. Can be any of the following:
@@ -644,16 +643,8 @@ FiftyOne can infer the appropriate values to use:
 
 -   **label_type**: if omitted, the |Label| type of the field will be used to
     infer the appropriate value for this parameter
--   **classes**: if omitted for a non-semantic segmentation field, the class
-    lists from the :meth:`classes <fiftyone.core.dataset.Dataset.classes>` or
-    :meth:`default_classes <fiftyone.core.dataset.Dataset.default_classes>`
-    properties of your dataset will be used, if available. Otherwise, the
-    observed labels on your dataset will be used to construct a classes list
--   **mask_targets**: if omitted for a semantic segmentation field, the mask
-    targets from the
-    :meth:`mask_targets <fiftyone.core.dataset.Dataset.mask_targets>` or
-    :meth:`default_mask_targets <fiftyone.core.dataset.Dataset.default_mask_targets>`
-    properties of your dataset will be used, if available
+-   **classes**: if omitted for a non-semantic segmentation field, the observed
+    labels on your dataset will be used to construct a classes list
 
 .. _annotation-label-attributes:
 

docs/source/user_guide/dataset_creation/datasets.rst

@@ -1553,16 +1553,16 @@ COCO format:
     dataset.export(
         export_dir="/tmp/coco",
         dataset_type=fo.types.COCODetectionDataset,
-        classes=classes,
         label_field="ground_truth",
+        classes=classes,
     )
 
     # Export predictions
     dataset.export(
         dataset_type=fo.types.COCODetectionDataset,
         labels_path="/tmp/coco/predictions.json",
-        classes=classes,
         label_field="predictions",
+        classes=classes,
     )
 
     # Now load ground truth labels into a new dataset
@@ -1577,7 +1577,7 @@ COCO format:
         dataset2,
         "predictions",
         "/tmp/coco/predictions.json",
-        classes=classes,
+        classes,
     )
 
     # Verify that ground truth and predictions were imported as expected
@@ -2147,16 +2147,16 @@ images-and-labels and labels-only data in YOLO format:
     dataset.export(
         export_dir="/tmp/yolov4",
         dataset_type=fo.types.YOLOv4Dataset,
-        classes=classes,
         label_field="ground_truth",
+        classes=classes,
     )
 
     # Export predictions
     dataset.export(
         dataset_type=fo.types.YOLOv4Dataset,
         labels_path="/tmp/yolov4/predictions",
-        classes=classes,
         label_field="predictions",
+        classes=classes,
     )
 
     # Now load ground truth labels into a new dataset
@@ -2171,7 +2171,7 @@ images-and-labels and labels-only data in YOLO format:
         dataset2,
         "predictions",
         "/tmp/yolov4/predictions",
-        classes=classes,
+        classes,
     )
 
     # Verify that ground truth and predictions were imported as expected
@@ -2327,16 +2327,16 @@ images-and-labels and labels-only data in YOLO format:
         export_dir="/tmp/yolov5",
         dataset_type=fo.types.YOLOv5Dataset,
         split="validation",
-        classes=classes,
         label_field="ground_truth",
+        classes=classes,
     )
 
     # Export predictions
     view.export(
         dataset_type=fo.types.YOLOv5Dataset,
         labels_path="/tmp/yolov5/predictions/validation",
-        classes=classes,
         label_field="predictions",
+        classes=classes,
     )
 
     # Now load ground truth labels into a new dataset
@@ -2352,7 +2352,7 @@ images-and-labels and labels-only data in YOLO format:
         dataset2,
         "predictions",
         "/tmp/yolov5/predictions/validation",
-        classes=classes,
+        classes,
     )
 
     # Verify that ground truth and predictions were imported as expected

docs/source/user_guide/dataset_creation/index.rst

@@ -369,16 +369,16 @@ predictions to take advantage of FiftyOne's
         dataset.export(
             export_dir="/tmp/coco",
             dataset_type=fo.types.COCODetectionDataset,
-            classes=classes,
             label_field="ground_truth",
+            classes=classes,
         )
 
         # Export predictions
         dataset.export(
             dataset_type=fo.types.COCODetectionDataset,
             labels_path="/tmp/coco/predictions.json",
-            classes=classes,
             label_field="predictions",
+            classes=classes,
         )
 
         # Now load ground truth labels into a new dataset
@@ -393,7 +393,7 @@ predictions to take advantage of FiftyOne's
             dataset2,
             "predictions",
             "/tmp/coco/predictions.json",
-            classes=classes,
+            classes,
         )
 
         # Verify that ground truth and predictions were imported as expected
@@ -432,16 +432,16 @@ predictions to take advantage of FiftyOne's
         dataset.export(
             export_dir="/tmp/yolov4",
             dataset_type=fo.types.YOLOv4Dataset,
-            classes=classes,
             label_field="ground_truth",
+            classes=classes,
         )
 
         # Export predictions
         dataset.export(
             dataset_type=fo.types.YOLOv4Dataset,
             labels_path="/tmp/yolov4/predictions",
-            classes=classes,
             label_field="predictions",
+            classes=classes,
         )
 
         # Now load ground truth labels into a new dataset
@@ -456,7 +456,7 @@ predictions to take advantage of FiftyOne's
             dataset2,
             "predictions",
             "/tmp/yolov4/predictions",
-            classes=classes,
+            classes,
         )
 
         # Verify that ground truth and predictions were imported as expected

docs/source/user_guide/export_datasets.rst

@@ -349,26 +349,16 @@ Class lists
 -----------
 
 Certain labeled image/video export formats such as
-:ref:`COCO <COCODetectionDataset-export>` or
+:ref:`COCO <COCODetectionDataset-export>` and
 :ref:`YOLO <YOLOv5Dataset-export>` store an explicit list of classes for the
 label field being exported.
 
-In such cases, all exporters provided by FiftyOne use the following strategy
-to retrieve the class list when exporting a given `label_field`:
+By convention, all exporters provided by FiftyOne should provide a `classes`
+parameter that allows for manually specifying the classes list to use.
 
-1.  If the exporter provides a parameter like `classes` that allows for
-    manually specifying the classes, this list is used
-2.  If the :meth:`classes <fiftyone.core.dataset.Dataset.classes>` dict of the
-    sample collection contains `label_field`, this class list is used
-3.  If the collection's
-    :meth:`default_classes <fiftyone.core.dataset.Dataset.default_classes>`
-    attribute is non-empty, this class list is used
-4.  If the collection's :meth:`info <fiftyone.core.dataset.Dataset.info>` dict
-    contains a class list under the `classes` key, this list is used
-
-If no explicit class list is available via the above methods, the observed
-classes in the collection being exported are used, which may be a subset of the
-classes in the parent dataset when exporting a view.
+If no explicit class list is provided, the observed classes in the collection
+being exported are used, which may be a subset of the classes in the parent
+dataset when exporting a view.
 
 .. note::
 
@@ -397,26 +387,17 @@ classes in the parent dataset when exporting a view.
     # Create a view that only contains cats and dogs
     view = dataset.filter_labels("ground_truth", F("label").is_in(["cat", "dog"]))
 
-    # By default, `default_classes` will be used to populate the COCO categories
+    # By default, only the observed classes will be stored as COCO categories
     view.export(
         labels_path="/tmp/coco1.json",
         dataset_type=fo.types.COCODetectionDataset,
     )
 
-    # But, since we're only interested in exporting cats and dogs, we can override
-    # this by manually providing the `classes` argument
+    # However, if desired, we can explicitly provide a classes list
     view.export(
         labels_path="/tmp/coco2.json",
         dataset_type=fo.types.COCODetectionDataset,
-        classes=["cat", "dog"],
-    )
-
-    # Equivalently, we can clear `default_classes` first so that the observed
-    # labels (only cats and dogs in this case) will be used
-    dataset.default_classes = None
-    view.export(
-        labels_path="/tmp/coco3.json",
-        dataset_type=fo.types.COCODetectionDataset,
+        classes=dataset.default_classes,
     )
 
 .. _supported-export-formats:

fiftyone/core/collections.py

@@ -2010,11 +2010,7 @@ def evaluate_classifications(
                 instances
             eval_key (None): a string key to use to refer to this evaluation
             classes (None): the list of possible classes. If not provided,
-                classes are loaded from
-                :meth:`fiftyone.core.dataset.Dataset.classes` or
-                :meth:`fiftyone.core.dataset.Dataset.default_classes` if
-                possible, or else the observed ground truth/predicted labels
-                are used
+                the observed ground truth/predicted labels are used
             missing (None): a missing label string. Any None-valued labels
                 are given this label for results purposes
             method ("simple"): a string specifying the evaluation method to use.
@@ -2116,11 +2112,7 @@ def evaluate_detections(
                 or :class:`fiftyone.core.labels.TemporalDetections`
             eval_key (None): a string key to use to refer to this evaluation
             classes (None): the list of possible classes. If not provided,
-                classes are loaded from
-                :meth:`fiftyone.core.dataset.Dataset.classes` or
-                :meth:`fiftyone.core.dataset.Dataset.default_classes` if
-                possible, or else the observed ground truth/predicted labels
-                are used
+                the observed ground truth/predicted labels are used
             missing (None): a missing label string. Any unmatched objects are
                 given this label for results purposes
             method (None): a string specifying the evaluation method to use.
@@ -2211,10 +2203,7 @@ def evaluate_segmentations(
                 instances
             eval_key (None): a string key to use to refer to this evaluation
             mask_targets (None): a dict mapping mask values to labels. If not
-                provided, mask targets are loaded from
-                :meth:`fiftyone.core.dataset.Dataset.mask_targets` or
-                :meth:`fiftyone.core.dataset.Dataset.default_mask_targets` if
-                possible, or else the observed pixel values are used
+                provided, the observed pixel values are used
             method ("simple"): a string specifying the evaluation method to
                 use. Supported values are ``("simple")``
             **kwargs: optional keyword arguments for the constructor of the

fiftyone/utils/annotations.py

@@ -115,10 +115,8 @@ def annotate(
             ``label_field`` or all fields in ``label_schema`` without classes
             specified. All new label fields must have a class list provided via
             one of the supported methods. For existing label fields, if classes
-            are not provided by this argument nor ``label_schema``, they are
-            parsed from
-            :meth:`fiftyone.core.collections.SampleCollection.get_classes` if
-            possible, or else the observed labels on your dataset are used
+            are not provided by this argument nor ``label_schema``, the
+            observed labels on your dataset are used
         attributes (True): specifies the label attributes of each label
             field to include (other than their ``label``, which is always
             included) in the annotation export. Can be any of the
@@ -772,10 +770,6 @@ def _get_classes(
             % label_field
         )
 
-    classes = samples.get_classes(label_field)
-    if classes:
-        return classes
-
     _, label_path = samples._get_label_field_path(label_field, "label")
     return sorted(
         set(samples._dataset.distinct(label_path))
@@ -815,12 +809,6 @@ def _get_mask_targets(samples, mask_targets, label_field, label_info):
     if "mask_targets" in label_info:
         mask_targets = label_info["mask_targets"]
 
-    if mask_targets is None and label_field in samples.mask_targets:
-        mask_targets = samples.mask_targets[label_field]
-
-    if mask_targets is None and samples.default_mask_targets:
-        mask_targets = samples.default_mask_targets
-
     if mask_targets is None:
         mask_targets = {i: str(i) for i in range(1, 256)}
         mask_targets[0] = "background"