Tag
This paper introduces a plug-and-play 2D motion interface that enables pretrained motion language models to process 2D inputs from monocular videos without retraining, improving real-world applicability.