companion dataset

README.md

@@ -81,6 +81,9 @@ Using RefOI, we evaluate several state-of-the-art VLMs and uncover **three tiers
 ![Overview](vlm-reg.png)
 
 
+For token-level annotation of referring expressions, see the companion dataset [RefOI-TLHF](https://huggingface.co/datasets/Seed42Lab/RefOI-TLHF), which provides minimal span supervision for both human- and model-generated descriptions.
+
+
 ## Dataset Schema and Split
 
 ### Data Fields