diff --git a/mmocr/datasets/icdar_dataset.py b/mmocr/datasets/icdar_dataset.py
index 83cceea8d..68fd911ad 100644
--- a/mmocr/datasets/icdar_dataset.py
+++ b/mmocr/datasets/icdar_dataset.py
@@ -39,7 +39,7 @@ class IcdarDataset(CocoDataset):
             None img. The maximum extra number of cycles to get a valid
             image. Defaults to 1000.
     """
-    METAINFO = {'CLASSES': ('text', )}
+    METAINFO = {'classes': ('text', )}
 
     def parse_data_info(self, raw_data_info: dict) -> Union[dict, List[dict]]:
         """Parse raw annotation to target format.
diff --git a/tests/test_datasets/test_icdar_dataset.py b/tests/test_datasets/test_icdar_dataset.py
index d4b3b5448..575c50619 100644
--- a/tests/test_datasets/test_icdar_dataset.py
+++ b/tests/test_datasets/test_icdar_dataset.py
@@ -115,7 +115,7 @@ def _create_dummy_icdar_json(self, json_name):
             'categories':
             categories
         }
-        self.metainfo = dict(CLASSES=('text'))
+        self.metainfo = dict(classes=('text'))
         mmengine.dump(fake_json, json_name)
 
     def test_icdar_dataset(self):
@@ -130,7 +130,7 @@ def test_icdar_dataset(self):
             data_prefix=dict(img_path='imgs'),
             metainfo=self.metainfo,
             pipeline=[])
-        self.assertEqual(dataset.metainfo['CLASSES'], self.metainfo['CLASSES'])
+        self.assertEqual(dataset.metainfo['classes'], self.metainfo['classes'])
         dataset.full_init()
         self.assertEqual(len(dataset), 2)
         self.assertEqual(len(dataset.load_data_list()), 2)