fix: resolve multiple validation bugs, add ontology auto-detection, fix documentation

crorry-dev · crorry-dev · commit d759fc0465a4 · 2026-04-08T11:53:29.000+02:00
- Fix ignore_tracks class with no attributes causing validation errors
- Fix three bugs in _check_attribute_scopes (identity comparison, early return, tracking granularity)
- Fix redundant sensor lookup, inconsistent horizon_tolerance default, bare ValueError/AssertionError
- Remove dead _build_error_path function, fix placeholder docstring
- Add detect_ontology() for automatic ontology detection based on unique class names
- Integrate auto-detection into validate() when no ontology explicitly provided
- Fix incorrect AutomatedTrain and OSDAR26 class listings in README, api.rst, and howtos
- Remove ignore_tracks from OSDAR26 unique classes (exists in both automatedtrain and osdar26)
diff --git a/README.md b/README.md
@@ -181,25 +181,29 @@ Pre-built ontology parameter files are provided and accessible via the API:
 |----------|------|---------|-------------|
 | `osdar23` | OSDAR23 | OSDAR23 | Original railway environment annotation ontology. Includes standard occlusion ranges (0-25%, 25-50%, 50-75%, 75-99%, 100%) and core railway classes. |
 | `osdar26` | OSDAR26 | OSDAR26 | Extended railway environment ontology with 25 object classes. Features comprehensive signal aspects (Hp, Ks, Vr, Zs, Sh variants), updated occlusion ranges (0-24%, 25-49%, 50-74%, 75-99%, 100%), and additional classes. |
-| `automatedtrain` | AutomatedTrain | AutomatedTrain | Specialized ontology for automated train perception and safety-critical annotation. Includes obstacle detection, platform recognition, level crossings, speed signs. |
+| `automatedtrain` | AutomatedTrain | AutomatedTrain | Specialized ontology for automated train perception. Extended class set including train_front, reflective_test_object, trampoline, plane, and ignore_tracks. |
 
 ### OSDAR26 Classes (25 total)
 - **Persons**: `person`, `crowd`
 - **Personal Mobility**: `personal_item`, `pram`, `scooter`, `wheelchair`
 - **Vehicles**: `bicycle`, `group_of_bicycles`, `motorcycle`, `road_vehicle`
 - **Animals**: `animal`, `group_of_animals`
 - **Railway Vehicles**: `train`, `wagon`, `drag_shoe`
-- **Track Infrastructure**: `track`, `transition`, `switch`
+- **Track Infrastructure**: `track`, `switch`, `ignore_tracks`
 - **Signaling**: `signal`, `signal_pole`, `signal_bridge`, `catenary_pole`, `buffer_stop`
 - **Hazards**: `flame`, `smoke`
 
-### AutomatedTrain Classes
+### AutomatedTrain Classes (29 total)
 
-The AutomatedTrain ontology supports safety-critical classes for automated train operation:
-- **Core Classes**: `person`, `crowd`, `train`, `wagon`, `track`, `transition`, `switch`
+- **Persons**: `person`, `crowd`
+- **Personal Mobility**: `personal_item`, `pram`, `scooter`, `wheelchair`
+- **Vehicles**: `bicycle`, `group_of_bicycles`, `motorcycle`, `road_vehicle`
+- **Animals**: `animal`, `group_of_animals`
+- **Railway Vehicles**: `train`, `train_front`, `wagons`, `drag_shoe`
+- **Track Infrastructure**: `track`, `switch`, `ignore_tracks`
 - **Signaling**: `signal`, `signal_pole`, `signal_bridge`, `catenary_pole`, `buffer_stop`
-- **Vehicles**: `road_vehicle`, `bicycle`, `motorcycle`, `animal`
-- **Safety-Critical**: `obstacle`, `platform`, `level_crossing`, `speed_sign`
+- **Hazards**: `flame`, `smoke`
+- **Special**: `reflective_test_object`, `trampoline`, `plane`
 
 ## Supported Sensors
 
diff --git a/config/ontologies/automatedtrain.yaml b/config/ontologies/automatedtrain.yaml
@@ -523,7 +523,7 @@ track:
     attribute_type: boolean
     scope: frame
 
-  trackId:
+  trackID:
     attribute_type: string
     scope: annotation
 
@@ -812,10 +812,6 @@ smoke:
         scope: annotation
 
 ignore_tracks:
-  noError:
-    attribute_type:
-      type: single-select
-      options: [noError]
 
 
 reflective_test_object:
diff --git a/config/schemas/ontology_schema_v2.yaml b/config/schemas/ontology_schema_v2.yaml
@@ -47,7 +47,7 @@ definitions:
     class:
         additionalProperties:
             "$ref": "#/definitions/attribute"
-        type: object
+        type: [object, "null"]
 
     integer_attribute:
         const: integer
diff --git a/docs/source/api.rst b/docs/source/api.rst
@@ -40,8 +40,8 @@ Available Ontologies
     and classes for personal_item, pram, scooter, flame, smoke.
 
 **AutomatedTrain** (``automatedtrain``)
-    Specialized ontology for automated train perception and safety-critical annotation.
-    Includes obstacle detection, platform recognition, level crossings, and speed signs.
+    Specialized ontology for automated train perception. Extended class set (29 classes)
+    including train_front, reflective_test_object, trampoline, plane, and ignore_tracks.
 
 **OSDAR23** (``osdar23``)
     Original railway environment annotation ontology for the OSDAR23 dataset.
diff --git a/docs/source/howtos/1 Validating Scenes.rst b/docs/source/howtos/1 Validating Scenes.rst
@@ -59,17 +59,15 @@ Pre-built ontology parameter files are provided in the ``config/ontologies/`` di
     - Updated occlusion ranges: 0-24%, 25-49%, 50-74%, 75-99%, 100%
     - Additional object classes: personal_item (suitcase, backpack, handbag, bag, etc.), pram (stroller, buggy, babySeat), scooter (eScooter, scooter, hoverboard)
     - Enhanced animal species coverage (dog, cat, racoon, badger, swan, sheep, cow, horse, pig, fox, wolf, wildBoar, deer, stork, rabbit, bird)
-    - Complete railway infrastructure elements (track, switch, transition, catenary_pole, signal_pole, buffer_stop)
+    - Complete railway infrastructure elements (track, switch, ignore_tracks, catenary_pole, signal_pole, buffer_stop)
     - Hazard detection (flame, smoke)
 
 **AutomatedTrain** (``config/ontologies/automatedtrain.yaml``)
-    Specialized ontology for automated train perception and safety-critical railway environment annotation:
+    Specialized ontology for automated train perception with 29 object classes:
 
-    - **Safety-critical classes**: obstacle, platform, level_crossing, speed_sign
+    - **Extended classes**: train_front, reflective_test_object, trampoline, plane, ignore_tracks
     - Switch state tracking (straight, diverging)
-    - Emergency vehicle detection
-    - Level crossing state monitoring (open, closing, closed, opening)
-    - Boolean safety indicators for critical situations
+    - Uses plural ``wagons`` (vs. singular ``wagon`` in OSDAR26)
 
 Usage Examples
 ##############
diff --git a/raillabel_providerkit/__init__.py b/raillabel_providerkit/__init__.py
@@ -10,6 +10,7 @@
 from .convert.convert import convert
 from .export.export_scenes import export_scenes
 from .ontologies import (
+    detect_ontology,
     get_ontology_path,
     get_schema_path,
     list_available_ontologies,
@@ -29,6 +30,7 @@
     "convert",
     "export_scenes",
     "validate",
+    "detect_ontology",
     "get_ontology_path",
     "get_schema_path",
     "list_available_ontologies",
diff --git a/raillabel_providerkit/__main__.py b/raillabel_providerkit/__main__.py
@@ -452,8 +452,8 @@ def export_command(
 @click.option(
     "--horizon-tolerance",
     type=float,
-    default=0.0,
-    help="Tolerance buffer as percentage above horizon (e.g., 5.0 for 5%%). Default is 0.0.",
+    default=10.0,
+    help="Tolerance buffer as percentage above horizon (e.g., 10.0 for 10%%). Default is 10.0.",
 )
 @click.option("-q", "--quiet", is_flag=True, help="Disable progress bars")
 def run_raillabel_providerkit(  # noqa: PLR0913
diff --git a/raillabel_providerkit/ontologies/__init__.py b/raillabel_providerkit/ontologies/__init__.py
@@ -4,13 +4,15 @@
 """Package for built-in ontology and schema definitions."""
 
 from .manager import (
+    detect_ontology,
     get_ontology_path,
     get_schema_path,
     list_available_ontologies,
     list_available_schemas,
 )
 
 __all__ = [
+    "detect_ontology",
     "get_ontology_path",
     "get_schema_path",
     "list_available_ontologies",
diff --git a/raillabel_providerkit/ontologies/manager.py b/raillabel_providerkit/ontologies/manager.py
@@ -14,8 +14,12 @@
 All ontology files are stored centrally in the config/ontologies/ directory.
 """
 
+from __future__ import annotations
+
 from pathlib import Path
 
+import raillabel
+
 
 def _get_config_dir() -> Path:
     """Get the path to the config directory.
@@ -156,7 +160,55 @@ def list_available_schemas() -> list[str]:
     return ["raillabel", "osdar23", "osdar26", "automatedtrain", "understand_ai_t4", "ontology"]
 
 
+_UNIQUE_CLASSES: dict[str, set[str]] = {
+    "automatedtrain": {"train_front", "reflective_test_object", "trampoline", "plane"},
+    "osdar23": {"transition"},
+}
+
+_EXCLUSIVE_PAIRS: dict[str, set[str]] = {
+    "automatedtrain": {"personal_item", "pram", "scooter"},
+    "osdar26": {"wagon"},
+}
+
+
+def detect_ontology(scene: raillabel.Scene) -> str | None:
+    """Detect which ontology matches the object types in a scene.
+
+    Uses unique class names to determine which ontology should be applied. Returns
+    None if no ontology can be reliably detected.
+
+    Parameters
+    ----------
+    scene : raillabel.Scene
+        The loaded scene to analyze.
+
+    Returns
+    -------
+    str | None
+        Name of the detected ontology ('automatedtrain', 'osdar23', 'osdar26'),
+        or None if detection is ambiguous or no objects are present.
+    """
+    object_types = {obj.type for obj in scene.objects.values()}
+
+    if not object_types:
+        return None
+
+    for ontology_name, unique_classes in _UNIQUE_CLASSES.items():
+        if object_types & unique_classes:
+            return ontology_name
+
+    if "wagon" in object_types:
+        return "osdar26"
+    if "wagons" in object_types:
+        if object_types & _EXCLUSIVE_PAIRS["automatedtrain"]:
+            return "automatedtrain"
+        return "osdar23"
+
+    return None
+
+
 __all__ = [
+    "detect_ontology",
     "get_ontology_path",
     "get_schema_path",
     "list_available_ontologies",
diff --git a/raillabel_providerkit/validation/validate.py b/raillabel_providerkit/validation/validate.py
@@ -9,6 +9,7 @@
 from raillabel import Scene
 from raillabel.json_format import JSONScene
 
+from raillabel_providerkit.ontologies import detect_ontology, get_ontology_path
 from raillabel_providerkit.validation import Issue
 
 from . import (
@@ -95,6 +96,10 @@ def validate(  # noqa: C901, PLR0912, PLR0913
 
     if ontology_source is not None:
         errors.extend(validate_ontology(scene, ontology_source))
+    else:
+        detected = detect_ontology(scene)
+        if detected is not None:
+            errors.extend(validate_ontology(scene, get_ontology_path(detected)))
 
     if validate_for_empty_frames:
         errors.extend(validate_empty_frames(scene))
diff --git a/raillabel_providerkit/validation/validate_annotation_type_per_sensor/validate_annotation_type_per_sensor.py b/raillabel_providerkit/validation/validate_annotation_type_per_sensor/validate_annotation_type_per_sensor.py
@@ -84,8 +84,6 @@ def _sensor_type(scene: raillabel.Scene, sensor_id: str) -> str | None:
     """Return the sensor type ('camera', 'lidar', 'radar') or None if unknown."""
     sensors = getattr(scene, "sensors", {}) or {}
     sensor = sensors.get(sensor_id)
-    if sensor is None and hasattr(sensors, "get"):
-        sensor = sensors.get(sensor_id)
     if sensor is None:
         return None
     s_type = getattr(sensor, "type", None)
diff --git a/raillabel_providerkit/validation/validate_horizon/validate_horizon.py b/raillabel_providerkit/validation/validate_horizon/validate_horizon.py
@@ -61,7 +61,8 @@ def validate_horizon(
     for frame_uid, frame in filtered_scene.frames.items():
         for annotation_uid, annotation in frame.annotations.items():
             if not isinstance(annotation, Poly2d):
-                raise AssertionError  # noqa: TRY004
+                msg = f"Expected Poly2d annotation but got {type(annotation).__name__}"
+                raise TypeError(msg)
 
             sensor_id = annotation.sensor_id
 
@@ -119,7 +120,7 @@ def _validate_annotation_for_horizon(
 
         # Apply tolerance buffer: move horizon line up by the tolerance percentage
         # The horizon_y is measured from the top of the image (Y=0 is top)
-        # Moving it "up" means making it smaller
+        # Moving it "up" means making it smaller (more permissive for tracks)
         # tolerance_percent% of horizon_y is subtracted
         horizon_y_with_buffer = horizon_y * (1.0 - horizon_tolerance_percent / 100.0)
 
diff --git a/raillabel_providerkit/validation/validate_ontology/_ontology_classes/_object_classes.py b/raillabel_providerkit/validation/validate_ontology/_ontology_classes/_object_classes.py
@@ -38,7 +38,7 @@ def _attribute_fromdict(cls, attribute: dict) -> _Attribute:
             if attribute_class.supports(attribute):
                 return attribute_class.fromdict(attribute)
 
-        raise ValueError
+        raise ValueError(f"No attribute class supports definition: {attribute}")
 
     def _check_undefined_attributes(
         self, annotation_metadata: _AnnotationWithMetadata
diff --git a/raillabel_providerkit/validation/validate_ontology/_ontology_classes/_ontology.py b/raillabel_providerkit/validation/validate_ontology/_ontology_classes/_ontology.py
@@ -21,7 +21,8 @@ class _Ontology:
     @classmethod
     def fromdict(cls, data: dict) -> _Ontology:
         return _Ontology(
-            {class_id: _ObjectClass.fromdict(class_) for class_id, class_ in data.items()}, []
+            {class_id: _ObjectClass.fromdict(class_ or {}) for class_id, class_ in data.items()},
+            [],
         )
 
     def check(self, scene: raillabel.Scene) -> list[Issue]:
@@ -56,7 +57,7 @@ def _check_attribute_scopes(
         self, annotations_with_metadata: list[_AnnotationWithMetadata]
     ) -> list[Issue]:
         errors: list[Issue] = []
-        checked_attributes_by_object_type: dict[str, list[str]] = {}
+        checked_keys: set[tuple[str, object, str]] = set()
 
         for annotation_with_metadata in annotations_with_metadata:
             object_type_name = annotation_with_metadata.object_type
@@ -66,8 +67,7 @@ def _check_attribute_scopes(
                 continue
             object_class = self.classes[object_type_name]
 
-            if object_type_name not in checked_attributes_by_object_type:
-                checked_attributes_by_object_type[object_type_name] = []
+            object_id = annotation_with_metadata.annotation.object_id
 
             for (
                 attribute_name,
@@ -77,19 +77,20 @@ def _check_attribute_scopes(
                 if attribute is None:
                     continue
 
-                if attribute_name in checked_attributes_by_object_type[object_type_name]:
+                key = (object_type_name, object_id, attribute_name)
+                if key in checked_keys:
                     continue
-                checked_attributes_by_object_type[object_type_name].append(attribute_name)
+                checked_keys.add(key)
 
                 for other_annotation_with_metadata in annotations_with_metadata:
                     if (
-                        other_annotation_with_metadata is annotations_with_metadata
+                        other_annotation_with_metadata is annotation_with_metadata
                         or other_annotation_with_metadata.object_type != object_type_name
                     ):
                         continue
 
                     if attribute_name not in other_annotation_with_metadata.annotation.attributes:
-                        return []
+                        continue
                     other_attribute_value = other_annotation_with_metadata.annotation.attributes[
                         attribute_name
                     ]
diff --git a/raillabel_providerkit/validation/validate_schema/validate_schema.py b/raillabel_providerkit/validation/validate_schema/validate_schema.py
@@ -17,7 +17,7 @@ def validate_schema(data: dict) -> list[Issue]:
     Parameters
     ----------
     data : dict
-        _description_
+        The scene data as a dictionary to validate against the raillabel schema.
 
     Returns
     -------
@@ -75,13 +75,6 @@ def _make_errors_readable(errors: ValidationError) -> list[Issue]:  # noqa: C901
     return readable_errors
 
 
-def _build_error_path(loc: list[str]) -> str:
-    path = "$"
-    for part in loc:
-        path += f".{part}"
-    return path
-
-
 def _convert_missing_error_to_issue(error: dict) -> Issue:
     return Issue(
         type=IssueType.SCHEMA,
diff --git a/tests/test_detect_ontology.py b/tests/test_detect_ontology.py
diff --git a/tests/validation/validate_ontology/test_object_classes.py b/tests/validation/validate_ontology/test_object_classes.py
diff --git a/tests/validation/validate_ontology/test_ontology.py b/tests/validation/validate_ontology/test_ontology.py
diff --git a/tests/validation/validate_ontology/test_validate_ontology_schema.py b/tests/validation/validate_ontology/test_validate_ontology_schema.py
