RedCaps: web-curated image-text data created by the people, for the people

TL;DR

RedCaps is a large-scale, high-quality image-text dataset from Reddit, enabling improved captioning and visual representation learning with minimal filtering, benefiting vision and language tasks.

Contribution

The paper introduces RedCaps, a novel dataset of 12 million image-text pairs from Reddit, created with minimal filtering and guided by curated subreddits, enhancing data quality and diversity.

Findings

Models trained on RedCaps generate human-preferred, rich captions.

RedCaps-trained models transfer effectively to various downstream tasks.

RedCaps reduces filtering complexity compared to search engine or HTML-based datasets.

Abstract

Large datasets of paired images and text have become increasingly popular for learning generic representations for vision and vision-and-language tasks. Such datasets have been built by querying search engines or collecting HTML alt-text -- since web data is noisy, they require complex filtering pipelines to maintain quality. We explore alternate data sources to collect high quality data with minimal filtering. We introduce RedCaps -- a large-scale dataset of 12M image-text pairs collected from Reddit. Images and captions from Reddit depict and describe a wide variety of objects and scenes. We collect data from a manually curated set of subreddits, which give coarse image labels and allow us to steer the dataset composition without labeling individual instances. We show that captioning models trained on RedCaps produce rich and varied captions preferred by humans, and learn visual representations that transfer to many downstream tasks.