1 paper
Haiwen Diao, Penghao Wu, Hanming Deng +55
Recent large vision-language models (VLMs) remain fundamentally constrained by a persistent dichotomy: understanding and generation are treated as distinct problems, leading to fra…