Frozen Transformers in Language Models Are Effective Visual Encoder Layers

arXiv:2310.12973v2 Announce Type: replace-cross
Abstract: This paper reveals that large language models (LLMs), despite being trained solely on textual data, are surprisingly strong encoders for purely visual tasks in the absence of language. Even more intriguingly, this can be achieved by a simple yet previously overlooked strategy -- employing a frozen transformer block from pre-trained LLMs as a constituent encoder layer to directly process visual tokens. Our work pushes the boundaries of leveraging LLMs for computer vision tasks, significantly departing …

arxiv cs.ai cs.cl cs.cv cs.lg encoder language language models transformers type visual