WIT: Wikipedia-based Image Text Dataset for Multimodal Multilingual Machine Learning

TL;DR

The WIT dataset is a large, multilingual, and diverse visio-linguistic dataset derived from Wikipedia, designed to advance multimodal and multilingual machine learning with its extensive image-text pairs and real-world complexity.

Contribution

This paper introduces WIT, the largest and most multilingual visio-linguistic dataset from Wikipedia, enabling improved pretraining and evaluation for multimodal multilingual models.

Findings

WIT is the largest multimodal dataset with 37.6 million examples.

WIT covers over 100 languages, each with at least 12K examples.

WIT provides a challenging real-world test set for image-text retrieval.

Abstract

The milestone improvements brought about by deep representation learning and pre-training techniques have led to large performance gains across downstream NLP, IR and Vision tasks. Multimodal modeling techniques aim to leverage large high-quality visio-linguistic datasets for learning complementary information (across image and text modalities). In this paper, we introduce the Wikipedia-based Image Text (WIT) Dataset (https://github.com/google-research-datasets/wit) to better facilitate multimodal, multilingual learning. WIT is composed of a curated set of 37.6 million entity rich image-text examples with 11.5 million unique images across 108 Wikipedia languages. Its size enables WIT to be used as a pretraining dataset for multimodal models, as we show when applied to downstream tasks such as image-text retrieval. WIT has four main and unique advantages. First, WIT is the largest multimodal dataset by the number of image-text examples by 3x (at the time of writing). Second, WIT is massively multilingual (first of its kind) with coverage over 100+ languages (each of which has at least 12K examples) and provides cross-lingual texts for many images. Third, WIT represents a more diverse set of concepts and real world entities relative to what previous datasets cover. Lastly, WIT provides a very challenging real-world test set, as we empirically illustrate using an image-text retrieval task as an example.