Ah yes, this model treats this purely as image captioning. The model isn't open source yet.