VLMo: Unified Vision-Language Pre-Training with Mixture-of-Modality-Experts

TL;DR

VLMo is a unified vision-language pretraining model that employs a novel Mixture-of-Modality-Experts Transformer, enabling flexible fine-tuning for various vision-language tasks and achieving state-of-the-art results.

Contribution

Introduction of MoME Transformer architecture and a stagewise pre-training strategy for unified vision-language modeling.

Findings

Achieves state-of-the-art results on VQA, NLVR2, and image-text retrieval.

Flexible use as fusion or dual encoder for different tasks.

Effective leveraging of large-scale image-only and text-only data.

Abstract

We present a unified Vision-Language pretrained Model (VLMo) that jointly learns a dual encoder and a fusion encoder with a modular Transformer network. Specifically, we introduce Mixture-of-Modality-Experts (MoME) Transformer, where each block contains a pool of modality-specific experts and a shared self-attention layer. Because of the modeling flexibility of MoME, pretrained VLMo can be fine-tuned as a fusion encoder for vision-language classification tasks, or used as a dual encoder for efficient image-text retrieval. Moreover, we propose a stagewise pre-training strategy, which effectively leverages large-scale image-only and text-only data besides image-text pairs. Experimental results show that VLMo achieves state-of-the-art results on various vision-language tasks, including VQA, NLVR2 and image-text retrieval. The code and pretrained models are available at https://aka.ms/vlmo.