diff --git a/benchmarks/set_matching_pytorch/train_sm.py b/benchmarks/set_matching_pytorch/train_sm.py
index 8257c20..8613534 100644
--- a/benchmarks/set_matching_pytorch/train_sm.py
+++ b/benchmarks/set_matching_pytorch/train_sm.py
@@ -29,7 +29,7 @@ def get_train_val_loader(
 ) -> Tuple[Any, Any]:
     label_dir_name = f"{train_year}-{valid_year}-split{split}"
 
-    iqon_outfits = IQONOutfits(root=root, split=split)
+    iqon_outfits = IQONOutfits(root=root)
 
     train, valid = iqon_outfits.get_trainval_data(label_dir_name)
     feature_dir = iqon_outfits.feature_dir
diff --git a/benchmarks/set_matching_pytorch/train_we.py b/benchmarks/set_matching_pytorch/train_we.py
index 1af9ac8..7cbf06f 100644
--- a/benchmarks/set_matching_pytorch/train_we.py
+++ b/benchmarks/set_matching_pytorch/train_we.py
@@ -33,7 +33,7 @@ def get_train_val_loader(
 ) -> Tuple[Any, Any]:
     label_dir_name = f"{train_year}-{valid_year}-split{split}"
 
-    iqon_outfits = IQONOutfits(root=root, split=split)
+    iqon_outfits = IQONOutfits(root=root)
 
     train, valid = iqon_outfits.get_trainval_data(label_dir_name)
     feature_dir = iqon_outfits.feature_dir
diff --git a/shift15m/datasets/outfitfeature.py b/shift15m/datasets/outfitfeature.py
index 7bccc21..af3ab5d 100644
--- a/shift15m/datasets/outfitfeature.py
+++ b/shift15m/datasets/outfitfeature.py
@@ -182,7 +182,7 @@ def __init__(
         self,
         root: str = C.ROOT,
         split: int = 0,
-    ) -> None:
+    ) -> None:  # not used
         self.root = pathlib.Path(root)
         self.root.mkdir(parents=True, exist_ok=True)
         if not (self.root / "iqon_outfits.json").exists():
@@ -191,8 +191,11 @@ def __init__(
 
         self._label_dir = self.root / "set_matching/labels"
         if not self._label_dir.exists():
+            print("Making train/val dataset.")
             self._label_dir.mkdir(parents=True, exist_ok=True)
-            self._make_trainval_dataset(seed=split)
+            splits = [0, 1, 2]
+            for _s in splits:
+                self._make_trainval_dataset(seed=_s)
 
         self._feature_dir = self.root / "features"
         if not self._feature_dir.exists():
@@ -231,8 +234,6 @@ def _make_trainval_dataset(
         min_like_num: int = 50,
         seed: int = 0,
     ):
-        print("Make train/val dataset.")
-
         np.random.seed(seed)
         num_train, num_val, num_test = 30816, 3851, 3851  # max size