diff --git a/vlmeval/dataset/__init__.py b/vlmeval/dataset/__init__.py
index 4d7439c91..539580fe2 100644
--- a/vlmeval/dataset/__init__.py
+++ b/vlmeval/dataset/__init__.py
@@ -103,6 +103,7 @@
 from .qbench_video import QBench_Video, QBench_Video_MCQ, QBench_Video_VQA
 from .reasonmap_plus import ReasonMap_Plus
 from .refcoco import RefCOCODataset
+from .refcoco_m import RefCOCOMDataset
 from .refspatial import RefSpatialDataset
 from .refspatialbench import RefSpatialBench
 from .robospatialbench import RoboSpatialBench
@@ -286,7 +287,7 @@ def evaluate(self, eval_file, **judge_kwargs):
     OmniEarthMCQBench, VisFactor, OSTDataset, OCRBench_v2, TreeBench, CVQA, M4Bench,
     AyaVisionBench, TopViewRS, VLMBias, MMHELIX, MedqbenchMCQDataset, MathCanvas, MMReason,
     MedqbenchPairedDescriptionDataset, MedqbenchCaptionDataset, ChartMuseum, ChartQAPro, ReasonMap_Plus,  # noqa: E501
-    olmOCRBench, OceanOCRBench, MATBench, VLRMBench, RefCOCODataset, RefSpatialDataset,
+    olmOCRBench, OceanOCRBench, MATBench, VLRMBench, RefCOCODataset, RefCOCOMDataset, RefSpatialDataset,
     ERQADataset, SimpleVQA, HiPhODataset, MaCBench,
     UniSVG, SArena, VLMsAreBiased, MMESCIDataset, CoreCognition, GroundingME,
     FoxBench, VTCBench, Asclepius, PlotQA, ChartX, ChartBench, ChartCapDataset, WorldVQA, PuzzleVQA, VisualPuzzles,  # noqa: E501
diff --git a/vlmeval/dataset/refcoco_m.py b/vlmeval/dataset/refcoco_m.py
new file mode 100644
index 000000000..6346974e2
--- /dev/null
+++ b/vlmeval/dataset/refcoco_m.py
@@ -0,0 +1,14 @@
+from .refcoco import RefCOCODataset
+
+
+class RefCOCOMDataset(RefCOCODataset):
+    TYPE = 'GROUNDING'
+    MODALITY = 'IMAGE'
+    DATASET_URL = {
+        'RefCOCO-M': ''
+    }
+    DATASET_MD5 = {}
+
+    @classmethod
+    def supported_datasets(cls):
+        return ['RefCOCO-M']