Fix h-label bug of missing parent labels in output (#2626)

* Fix h-label bug of missing parent labels in output

* Fix h-label test data label schema

* Update CHANGELOG.md

---------
Signed-off-by: Songki Choi <[email protected]>

Author: Songki Choi

CHANGELOG.md

@@ -17,6 +17,7 @@ All notable changes to this project will be documented in this file.
 - Fix h-label loss normalization issue w/ exclusive label group of singe label (<https://github.com/openvinotoolkit/training_extensions/pull/2604>)
 - Fix division by zero in class incremental learning for classification (<https://github.com/openvinotoolkit/training_extensions/pull/2606>)
 - Fix saliency maps calculation issue for detection models (<https://github.com/openvinotoolkit/training_extensions/pull/2609>)
+- Fix h-label bug of missing parent labels in output (<https://github.com/openvinotoolkit/training_extensions/pull/2626>)
 
 ## \[v1.4.3\]
 

src/otx/algorithms/classification/adapters/mmcls/datasets/otx_datasets.py

@@ -300,6 +300,7 @@ class OTXHierarchicalClsDataset(OTXMultilabelClsDataset):
 
     def __init__(self, **kwargs):
         self.hierarchical_info = kwargs.pop("hierarchical_info", None)
+        self.label_schema = kwargs.pop("label_schema", None)
         super().__init__(**kwargs)
 
     def load_annotations(self):
@@ -308,6 +309,13 @@ def load_annotations(self):
         for i, _ in enumerate(self.otx_dataset):
             class_indices = []
             item_labels = self.otx_dataset[i].get_roi_labels(self.labels, include_empty=include_empty)
+            if self.label_schema:
+                # NOTE: Parent labels might be missing in annotations.
+                # This code fills the gap just in case.
+                full_item_labels = set()
+                for label in item_labels:
+                    full_item_labels.update(self.label_schema.get_ancestors(label))
+                item_labels = full_item_labels
             ignored_labels = self.otx_dataset[i].ignored_labels
             if item_labels:
                 num_cls_heads = self.hierarchical_info["num_multiclass_heads"]

src/otx/algorithms/classification/adapters/mmcls/task.py

@@ -190,6 +190,7 @@ def configure(
         elif self._hierarchical:
             options_for_patch_datasets["type"] = "OTXHierarchicalClsDataset"
             options_for_patch_datasets["hierarchical_info"] = self._hierarchical_info
+            options_for_patch_datasets["label_schema"] = self._task_environment.label_schema
             options_for_patch_evaluation["task"] = "hierarchical"
         elif self._selfsl:
             options_for_patch_datasets["type"] = "SelfSLDataset"

tests/assets/datumaro_h-label/annotations/train.json

@@ -3,6 +3,11 @@
   "categories": {
     "label": {
       "label_groups": [
+        {
+          "name": "shape",
+          "group_type": "exclusive",
+          "labels": ["blue", "green"]
+        },
         {
           "name": "blue",
           "group_type": "exclusive",

tests/assets/datumaro_h-label/annotations/valid.json

@@ -3,6 +3,11 @@
   "categories": {
     "label": {
       "label_groups": [
+        {
+          "name": "shape",
+          "group_type": "exclusive",
+          "labels": ["blue", "green"]
+        },
         {
           "name": "blue",
           "group_type": "exclusive",

tests/assets/datumaro_h-label_class_decremental/annotations/train.json

@@ -3,6 +3,11 @@
   "categories": {
     "label": {
       "label_groups": [
+        {
+          "name": "shape",
+          "group_type": "exclusive",
+          "labels": ["blue", "green"]
+        },
         {
           "name": "blue",
           "group_type": "exclusive",

tests/assets/datumaro_h-label_class_decremental/annotations/valid.json

@@ -3,6 +3,11 @@
   "categories": {
     "label": {
       "label_groups": [
+        {
+          "name": "shape",
+          "group_type": "exclusive",
+          "labels": ["blue", "green"]
+        },
         {
           "name": "blue",
           "group_type": "exclusive",