diff --git a/lerobot/common/datasets/push_dataset_to_hub/compute_stats.py b/lerobot/common/datasets/push_dataset_to_hub/compute_stats.py
index b03a17d9..9417d1dd 100644
--- a/lerobot/common/datasets/push_dataset_to_hub/compute_stats.py
+++ b/lerobot/common/datasets/push_dataset_to_hub/compute_stats.py
@@ -19,7 +19,7 @@ def get_stats_einops_patterns(dataset: LeRobotDataset | datasets.Dataset):
 
     dataloader = torch.utils.data.DataLoader(
         dataset,
-        num_workers=0,
+        num_workers=1,  # do not set to 0 when using `load_from_videos`
         batch_size=2,
         shuffle=False,
     )
diff --git a/lerobot/common/datasets/video_utils.py b/lerobot/common/datasets/video_utils.py
index 7524e881..2b82b3cb 100644
--- a/lerobot/common/datasets/video_utils.py
+++ b/lerobot/common/datasets/video_utils.py
@@ -13,6 +13,11 @@ from datasets.features.features import register_feature
 def load_from_videos(
     item: dict[str, torch.Tensor], video_frame_keys: list[str], videos_dir: Path, tolerance_s: float
 ):
+    """Note: When using data workers (e.g. DataLoader with num_workers>0), do not call this function
+    in the main process (e.g. by using a second Dataloader with num_workers=0). It will result in a Segmentation Fault.
+    This probably happens because a memory reference to the video loader is created in the main process and a
+    subprocess fails to access it.
+    """
     # since video path already contains "videos" (e.g. videos_dir="data/videos", path="videos/episode_0.mp4")
     data_dir = videos_dir.parent