i-Code V2: An Autoregressive Generation Framework over Vision, Language, and Speech Data

TL;DR

i-Code V2 introduces a pioneering autoregressive model capable of generating natural language from combined vision, language, and speech data, advancing multimodal AI by integrating diverse modalities into a unified generative framework.

Contribution

It is the first model to generate language from any combination of vision, language, and speech data, using a novel modality-fusing encoder and end-to-end pretraining across multiple modalities.

Findings

Outperforms state-of-the-art on 7 multimodal tasks

Effectively integrates multiple modalities into a shared representation

Demonstrates strong generalization across diverse multimodal signals

Abstract

The convergence of text, visual, and audio data is a key step towards human-like artificial intelligence, however the current Vision-Language-Speech landscape is dominated by encoder-only models which lack generative abilities. We propose closing this gap with i-Code V2, the first model capable of generating natural language from any combination of Vision, Language, and Speech data. i-Code V2 is an integrative system that leverages state-of-the-art single-modality encoders, combining their outputs with a new modality-fusing encoder in order to flexibly project combinations of modalities into a shared representational space. Next, language tokens are generated from these representations via an autoregressive decoder. The whole framework is pretrained end-to-end on a large collection of dual- and single-modality datasets using a novel text completion objective that can be generalized across arbitrary combinations of modalities. i-Code V2 matches or outperforms state-of-the-art single- and dual-modality baselines on 7 multimodal tasks, demonstrating the power of generative multimodal pretraining across a diversity of tasks and signals.