Multi-modal Auto-regressive Modeling via Visual Words

TL;DR

This paper introduces a novel multi-modal auto-regressive modeling approach using visual tokens, enabling LMMs to incorporate visual information with supervised labels, demonstrated through strong results on VQA tasks.

Contribution

It proposes the concept of visual tokens for discrete visual representation in LMMs, a first in unified multi-modal auto-regressive modeling.

Findings

Effective on 5 VQA tasks.

Outperforms existing methods on benchmark toolkits.

Validates the use of text embeddings for visual info.

Abstract

Large Language Models (LLMs), benefiting from the auto-regressive modelling approach performed on massive unannotated texts corpora, demonstrates powerful perceptual and reasoning capabilities. However, as for extending auto-regressive modelling to multi-modal scenarios to build Large Multi-modal Models (LMMs), there lies a great difficulty that the image information is processed in the LMM as continuous visual embeddings, which cannot obtain discrete supervised labels for classification.In this paper, we successfully perform multi-modal auto-regressive modeling with a unified objective for the first time.Specifically, we propose the concept of visual tokens, which maps the visual features to probability distributions over LLM's vocabulary, providing supervision information for visual modelling.We further explore the distribution of visual features in the semantic space within LMM and the possibility of using text embeddings to represent visual information.Experimental results and ablation studies on 5 VQA tasks and 4 benchmark toolkits validate the powerful performance of our proposed approach.