RefHCM: A Unified Model for Referring Perceptions in Human-Centric Scenarios
Published in IEEE Transactions on Multimedia (TMM), 2025
We introduce Referring Human Perceptions and propose RefHCM, a unified framework that employs sequence mergers to convert raw multimodal data into semantic tokens, enabling diverse human-centric referring tasks to be reformulated into a sequence-to-sequence paradigm. RefHCM effectively facilitates knowledge transfer across tasks and exhibits capabilities in handling complex reasoning.
| Paper | Code |
Recommended citation: Jie Huang, Ruibing Hou, Jiahe Zhao, Hong Chang, Shiguang Shan. (2025). "RefHCM: A Unified Model for Referring Perceptions in Human-Centric Scenarios." IEEE Transactions on Multimedia.
Download Paper