diff --git a/benchmark/camelyon/pure_substrafl/register_assets.py b/benchmark/camelyon/pure_substrafl/register_assets.py
index ec48721c..e0659243 100644
--- a/benchmark/camelyon/pure_substrafl/register_assets.py
+++ b/benchmark/camelyon/pure_substrafl/register_assets.py
@@ -98,7 +98,7 @@ def add_duplicated_dataset(
     data_sample_folder: os.PathLike,
     asset_keys: dict,
     msp_id: str,
-    kind: str = TaskType.TRAIN,
+    kind: str = TaskType.TRAIN.value,
 ) -> dict:
     """Update asset_keys.msp_id.{kind}_data_sample_keys so there is exactly `nb_data_sample` keys
     by adding the `data_sample_folder` as a data sample with the provided `client` as many times as it's need
@@ -126,7 +126,7 @@ def add_duplicated_dataset(
                 }
         msp_id (str): asset_keys key where to find the registered assets for the given client
         kind (str, optional): Kind of data sample to add, either train or test.  Defaults to
-        TaskType.TRAIN.
+        TaskType.TRAIN.value.
 
     Returns:
         dict: The updated asset_keys.
@@ -193,7 +193,7 @@ def get_train_data_nodes(
             data_sample_folder=train_folder,
             asset_keys=asset_keys,
             msp_id=msp_id,
-            kind=TaskType.TRAIN,
+            kind=TaskType.TRAIN.value,
         )
 
         train_data_nodes.append(