Vision as Unified Multimodal Generation

Xiaoyang Han, Jianhua Li, Kewang Deng +14

We formulate computer vision as unified multimodal generation, where heterogeneous visual tasks are expressed in the native text and image generation spaces of a unified multimodal…