ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Tasks

TLDR

ViLBERT is a Vision‑and‑Language BERT model that learns task‑agnostic joint representations of images and natural language. It extends BERT to a two‑stream architecture with co‑attentional transformer layers, pretrained on the Conceptual Captions dataset via two proxy tasks, and fine‑tuned for multiple vision‑and‑language tasks with only minor architectural changes. ViLBERT attains state‑of‑the‑art results on visual question answering, visual commonsense reasoning, referring expressions, and image retrieval, outperforming task‑specific models and showing that visual grounding can be pretrained and transferred.

Abstract

We present ViLBERT (short for Vision-and-Language BERT), a model for learning task-agnostic joint representations of image content and natural language. We extend the popular BERT architecture to a multi-modal two-stream model, processing both visual and textual inputs in separate streams that interact through co-attentional transformer layers. We pretrain our model through two proxy tasks on the large, automatically collected Conceptual Captions dataset and then transfer it to multiple established vision-and-language tasks -- visual question answering, visual commonsense reasoning, referring expressions, and caption-based image retrieval -- by making only minor additions to the base architecture. We observe significant improvements across tasks compared to existing task-specific models -- achieving state-of-the-art on all four tasks. Our work represents a shift away from learning groundings between vision and language only as part of task training and towards treating visual grounding as a pretrainable and transferable capability.