About the multi-modal completion.

Thanks for your excellent work.  I have a question after reading your paper.  In the paper, does a partial correspond to a GT for supervision? If so, why can multi-modal completion be achieved? :-)