Learning Audio-Video Modalities from Image Captions

TL;DR

This paper introduces a novel pipeline for transferring image captions to video clips, creating a large-scale weakly labeled audio-video dataset that improves retrieval and captioning performance with less data.

Contribution

The authors propose a new video mining pipeline that leverages image caption datasets to generate large-scale weakly labeled audio-video data, enabling effective multimodal training.

Findings

Achieves competitive video retrieval and captioning performance with 20x fewer clips than previous methods.

State-of-the-art results in audio retrieval using mined clips for pretraining.

Demonstrates the effectiveness of transferred image captions for multimodal learning.

Abstract

A major challenge in text-video and text-audio retrieval is the lack of large-scale training data. This is unlike image-captioning, where datasets are in the order of millions of samples. To close this gap we propose a new video mining pipeline which involves transferring captions from image captioning datasets to video clips with no additional manual effort. Using this pipeline, we create a new large-scale, weakly labelled audio-video captioning dataset consisting of millions of paired clips and captions. We show that training a multimodal transformed based model on this data achieves competitive performance on video retrieval and video captioning, matching or even outperforming HowTo100M pretraining with 20x fewer clips. We also show that our mined clips are suitable for text-audio pretraining, and achieve state of the art results for the task of audio retrieval.