Added automatic clipping for normalized annotations

Author: klemen1999

luxonis_ml/data/datasets/annotation.py

@@ -1,4 +1,5 @@
 import json
+import logging
 from abc import ABC, abstractmethod
 from datetime import datetime
 from typing import Any, ClassVar, Dict, List, Literal, Optional, Tuple, TypedDict, Union
@@ -20,6 +21,8 @@
 
 from ..utils.enums import LabelType
 
+logger = logging.getLogger(__name__)
+
 KeypointVisibility: TypeAlias = Literal[0, 1, 2]
 NormalizedFloat: TypeAlias = Annotated[float, Field(ge=0, le=1)]
 """C{NormalizedFloat} is a float that is restricted to the range [0, 1]."""
@@ -137,6 +140,20 @@ class BBoxAnnotation(Annotation):
 
     _label_type = LabelType.BOUNDINGBOX
 
+    @model_validator(mode="before")
+    @classmethod
+    def validate_values(cls, values: Dict[str, Any]) -> Dict[str, Any]:
+        warn = False
+        for key in ["x", "y", "w", "h"]:
+            if not (0 <= values[key] <= 1):
+                warn = True
+                values[key] = max(0, min(1, values[key]))
+        if warn:
+            logger.warning(
+                "BBox annotation has values outside of [0,1] range. Clipping them to [0,1]."
+            )
+        return values
+
     def to_numpy(self, class_mapping: Dict[str, int]) -> np.ndarray:
         class_ = class_mapping.get(self.class_, 0)
         return np.array([class_, self.x, self.y, self.w, self.h])
@@ -170,6 +187,26 @@ class KeypointAnnotation(Annotation):
 
     _label_type = LabelType.KEYPOINTS
 
+    @model_validator(mode="before")
+    @classmethod
+    def validate_values(cls, values: Dict[str, Any]) -> Dict[str, Any]:
+        warn = False
+        for i, keypoint in enumerate(values["keypoints"]):
+            new_keypoint = list(keypoint)
+            if not (0 <= keypoint[0] <= 1):
+                new_keypoint[0] = max(0, min(1, keypoint[0]))
+                warn = True
+            if not (0 <= keypoint[1] <= 1):
+                new_keypoint[1] = max(0, min(1, keypoint[1]))
+                warn = True
+            values["keypoints"][i] = tuple(new_keypoint)
+
+        if warn:
+            logger.warning(
+                "Keypoint annotation has values outside of [0,1] range. Clipping them to [0,1]."
+            )
+        return values
+
     def to_numpy(self, class_mapping: Dict[str, int]) -> np.ndarray:
         class_ = class_mapping.get(self.class_, 0)
         kps = np.array(self.keypoints).reshape((-1, 3)).astype(np.float32)
@@ -340,6 +377,26 @@ class PolylineSegmentationAnnotation(SegmentationAnnotation):
 
     points: List[Tuple[NormalizedFloat, NormalizedFloat]] = Field(min_length=3)
 
+    @model_validator(mode="before")
+    @classmethod
+    def validate_values(cls, values: Dict[str, Any]) -> Dict[str, Any]:
+        warn = False
+        for i, point in enumerate(values["points"]):
+            new_point = list(point)
+            if not (0 <= point[0] <= 1):
+                new_point[0] = max(0, min(1, point[0]))
+                warn = True
+            if not (0 <= point[1] <= 1):
+                new_point[1] = max(0, min(1, point[1]))
+                warn = True
+            values["points"][i] = tuple(new_point)
+
+        if warn:
+            logger.warning(
+                "Polyline annotation has values outside of [0,1] range. Clipping them to [0,1]."
+            )
+        return values
+
     def to_numpy(self, _: Dict[str, int], width: int, height: int) -> np.ndarray:
         polyline = [(round(x * width), round(y * height)) for x, y in self.points]
         mask = Image.new("L", (width, height), 0)