fix: abstract away category indexing differences (#254)

* fix: abstract away category indexing differences

Semseg and non-semseg datasets behave differently (position-based vs. explicit indexing).
Now computing index field in case position-based indexing is used.

Signed-off-by: Max SCHMELLER <max.schmeller@tier4.jp>

* Cherry-pick of b91ece3 "refactor(compatibility): move to schema package
"

Signed-off-by: Max SCHMELLER <max.schmeller@tier4.jp>

---------

Signed-off-by: Max SCHMELLER <max.schmeller@tier4.jp>

Author: mojomex

t4_devkit/helper/rendering.py

@@ -54,7 +54,7 @@ def __init__(self, t4: Tier4) -> None:
         """
         self._t4 = t4
         self._label2id: dict[str, int] = {
-            category.name: idx for idx, category in enumerate(self._t4.category)
+            category.name: category.index for category in self._t4.category
         }
 
         self._executor = concurrent.futures.ThreadPoolExecutor()

t4_devkit/schema/compatibility.py

@@ -0,0 +1,33 @@
+"""Compatibility helpers that abstract away differences between T4Dataset revisions."""
+
+from __future__ import annotations
+
+from t4_devkit.schema import Category
+
+
+def fix_category_table(categories: list[Category]) -> list[Category]:
+    """Fix unpopulated index fields in some T4Dataset revisions.
+
+    This function behaves differently in the below three cases:
+
+    - All `index` fields are set: the list is returned unmodified.
+    - All `index` fields are `None`: the position of each category in the list is used to compute
+      `index`. The resulting indices start at `0`.
+    - Some `index` fields are set and some are `None`: raise a `ValueError`.
+
+    Args:
+        categories (list[Category]): List of categories to fix.
+
+    Returns:
+        list[Category]: Fixed list of categories.
+
+    Raises:
+        ValueError: If the `index` field is set for some categories and `None` for others.
+    """
+    if any(category.index is None for category in categories):
+        if not all(category.index is None for category in categories):
+            raise ValueError("Category index is not set for some categories.")
+
+        for idx, category in enumerate(categories):
+            category.index = idx
+    return categories

t4_devkit/tier4.py

@@ -15,6 +15,7 @@
 from t4_devkit.dataclass import Box2D, Box3D, SemanticLabel, Shape, ShapeType
 from t4_devkit.helper import RenderingHelper, TimeseriesHelper
 from t4_devkit.schema import SchemaName, SensorModality, VisibilityLevel, build_schema
+from t4_devkit.schema.compatibility import fix_category_table
 
 if TYPE_CHECKING:
     from t4_devkit.typing import CameraIntrinsicLike, Vector3
@@ -181,6 +182,8 @@ def __init__(
             self.annotation_dir, SchemaName.CALIBRATED_SENSOR
         )
         self.category: list[Category] = load_table(self.annotation_dir, SchemaName.CATEGORY)
+        self.category = fix_category_table(self.category)
+
         self.ego_pose: list[EgoPose] = load_table(self.annotation_dir, SchemaName.EGO_POSE)
         self.instance: list[Instance] = load_table(self.annotation_dir, SchemaName.INSTANCE)
         self.keypoint: list[Keypoint] = load_table(self.annotation_dir, SchemaName.KEYPOINT)
@@ -265,7 +268,7 @@ def __make_reverse_index__(self, verbose: bool) -> None:
         }
 
         self._label2id: dict[str, int] = {
-            category.name: idx for idx, category in enumerate(self.category)
+            category.name: category.index for category in self.category
         }
 
         # add shortcuts