diff --git a/src/depth_anything_3/model/dinov2/vision_transformer.py b/src/depth_anything_3/model/dinov2/vision_transformer.py
index f0f8e23a..82cc49fa 100644
--- a/src/depth_anything_3/model/dinov2/vision_transformer.py
+++ b/src/depth_anything_3/model/dinov2/vision_transformer.py
@@ -385,7 +385,7 @@ def get_intermediate_layers(
         elif outputs[0][1].shape[-1] == (self.embed_dim * 2):
             outputs = [
                 torch.cat(
-                    [out[1][..., : self.embed_dim], self.norm(out[1][..., self.embed_dim :])],
+                    [self.norm(out[1][..., : self.embed_dim]), self.norm(out[1][..., self.embed_dim :])],
                     dim=-1,
                 )
                 for out in outputs