Chun Yuan

PaDoc: Layout-Grounded Parallel Decoding for Document Parsing

Hao Yu, Jiabo Zhan, Kang Liu, et al.

Closed-Form Spectral Regularization for Multi-Task Model Merging

Yongxian Wei, Runxi Cheng, Xingxuan Zhang, et al.

PromptHub: Enhancing Multi-Prompt Visual In-Context Learning with Locality-Aware Fusion, Concentration and Alignment

Tianci Luo, Jinpeng Wang, Shiyu Qin, et al.

PromptHub is a framework that improves visual in-context learning (learning vision tasks from image examples) by better combining multiple demonstration examples through spatial awareness and smarter training objectives. Unlike previous methods that treat image patches independently, PromptHub uses location information to understand context better and employs multiple complementary training goals to help the model learn more effectively. The approach shows strong performance across multiple vision tasks and works reliably even with different types of images.