diff --git a/tests/test_omnimat_data_url.py b/tests/test_omnimat_data_url.py
new file mode 100644
index 000000000..64978e625
--- /dev/null
+++ b/tests/test_omnimat_data_url.py
@@ -0,0 +1,37 @@
+import pandas as pd
+
+from vlmeval.dataset import omnimat as omnimat_module
+from vlmeval.dataset.omnimat import OmniMat
+
+EXPECTED_URLS = {
+    'OmniMat_QA': (
+        'https://huggingface.co/datasets/'
+        'Summer12138/OmniMat1K-VLMEvalKit/resolve/main/'
+        'vlmevalkit/OmniMat_QA.tsv'
+    ),
+    'OmniMat_CAL': (
+        'https://huggingface.co/datasets/'
+        'Summer12138/OmniMat1K-VLMEvalKit/resolve/main/'
+        'vlmevalkit/OmniMat_CAL.tsv'
+    ),
+}
+
+
+def test_omnimat_uses_published_tsv_urls():
+    assert OmniMat.DATASET_URL == EXPECTED_URLS
+
+
+def test_omnimat_downloads_published_tsv_when_local_data_is_missing(tmp_path, monkeypatch):
+    monkeypatch.setattr(omnimat_module, 'LMUDataRoot', lambda: str(tmp_path))
+    dataset = object.__new__(OmniMat)
+    downloaded = pd.DataFrame(
+        [{'index': '1', 'category_id': '1', 'id': '2', 'question': 'test'}]
+    )
+    called = []
+    dataset.prepare_tsv = lambda url: called.append(url) or downloaded.copy()
+
+    data = dataset.load_data('OmniMat_QA')
+
+    assert called == [EXPECTED_URLS['OmniMat_QA']]
+    assert data.iloc[0]['category_id'] == '01'
+    assert data.iloc[0]['id'] == '002'
diff --git a/vlmeval/dataset/omnimat.py b/vlmeval/dataset/omnimat.py
index 4e3e3892c..c2037140a 100644
--- a/vlmeval/dataset/omnimat.py
+++ b/vlmeval/dataset/omnimat.py
@@ -22,8 +22,16 @@ class OmniMat(ImageBaseDataset):
     DEFAULT_JUDGE = 'gemini-2.5-flash'
 
     DATASET_URL = {
-        'OmniMat_QA': '',
-        'OmniMat_CAL': '',
+        'OmniMat_QA': (
+            'https://huggingface.co/datasets/'
+            'Summer12138/OmniMat1K-VLMEvalKit/resolve/main/'
+            'vlmevalkit/OmniMat_QA.tsv'
+        ),
+        'OmniMat_CAL': (
+            'https://huggingface.co/datasets/'
+            'Summer12138/OmniMat1K-VLMEvalKit/resolve/main/'
+            'vlmevalkit/OmniMat_CAL.tsv'
+        ),
     }
 
     QA_PROMPT = (
@@ -66,6 +74,16 @@ def load_data(self, dataset):
                 _normalize_ids(data)
                 return data
 
+        url = self.DATASET_URL.get(dataset)
+        if url:
+            data = self.prepare_tsv(url)
+            if 'image' in data:
+                data['image'] = data['image'].fillna('')
+            if 'image_path' in data:
+                data['image_path'] = data['image_path'].fillna('')
+            _normalize_ids(data)
+            return data
+
         raise FileNotFoundError(
             f'{dataset}.tsv was not found. Run `python scripts/convert_omnimat.py` '
             'from the repository root first, or place the TSV under LMUData.'