We propose an efficient method to ground pretrained text-only language models to the visual domain, enabling them to process arbitrarily interleaved image-and-text data, and generate text interleaved with retrieved images. Our method leverages the abilities of language models learnt from large scale text-only pretraining, such as in-context learning and free-form text generation. We keep the language model frozen, and finetune input and output linear layers to enable cross-modality interactions. This allows our model to process arbitrarily interleaved image-and-text inputs, and generate free-form text interleaved with retrieved images. We achieve strong zero-shot performance on grounded tasks such as contextual image retrieval and multimodal dialogue, and showcase compelling interactive abilities. Our approach works with any off-the-shelf language model and paves the way towards an effective, general solution for leveraging pretrained language models in visually grounded settings.
@article{arxiv.2301.13823,
title = {Grounding Language Models to Images for Multimodal Inputs and Outputs},
author = {Jing Yu Koh and Ruslan Salakhutdinov and Daniel Fried},
journal= {arXiv preprint arXiv:2301.13823},
year = {2023}
}
Comments
Published in ICML 2023. Project page: https://jykoh.com/fromage