Unsupervised Vision-and-Language Pre-training Without Parallel Images and Captions

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5339–5350 June 6–11, 2021. ©2021 Association for Computational Linguistics 5339 Unsupervised Vision-and-Language Pre-training Without Parallel Images and Captions Liunian Harold Li†, Haoxuan You∗◦, Zhecan Wang∗◦, Alireza Zareian◦, Shih-Fu Chang◦ & Kai-Wei Chang† †University of California, Los Angeles ◦Columbia University liunian.harold.li@cs.ucla.edu, {hy2612,zw2627,az2407,sc250}@columbia.edu, kwchang@cs.ucla.edu Abstract Pre-trained contextual vision-and-language (V&L) models have achieved impressive per- formance on various benchmarks. However, existing models require a large amount of parallel image-caption data for pre-training. Such data are costly to collect and require cumbersome curation. Inspired by unsu- pervised machine translation, we investigate if a strong V&L representation model can be learned through unsupervised pre-training without image-caption corpora. In particu- lar, we propose to conduct “mask-and-predict” pre-training on text-only and image-only cor- pora and introduce the object tags detected by an object recognition model as anchor points to bridge two modalities. We find that such a simple approach achieves performance close to a model pre-trained with aligned data, on four English V&L benchmarks. Our work challenges the widely held notion that aligned data is necessary for V&L pre-training, while significantly reducing the amount of supervi- sion needed for V&L models. 1 Introduction Pre-trained contextual vision-and-language (V&L) models (Lu et al., 2019; Tan and Bansal, 2019; Li et al., 2019; Su et al., 2019; Chen et al., 2020c) have achieved high performance on various V&L tasks. However, different from contextual language models, such as BERT (Devlin et al., 2019a), which are trained on easily-accessible unannotated text corpora, existing V&L models are still a step away from self-supervision. They require a massive amount of aligned text-image pairs for “mask-and-predict” pre-training. Such aligned data are costly to collect and hard to scale up. For example, the widely used MS-COCO dataset (Chen et al., 2015) requires extensive ∗The two authors contributed equally. annotation from crowd workers.1 In this paper, we explore unsupervised V&L pre-training with unaligned image and text cor- pora.2 This research direction aligns with the theme of unsupervised and self-supervised learning that moves from heavily-annotated data to unannotated data, e.g. unsupervised machine translation (Lam- ple et al., 2018) and unsupervised image caption- ing (Feng et al., 2019). Unsupervised V&L pre- training is highly desirable as in many domains, aligned data is scarce (e.g. multimodal hate speech detection (Kiela et al., 2020) and the medical do- main (Li et al., 2020c)) and it is easier to collect un- aligned text and images. In addition to its practical implication, our endeavour challenges the widely held notion that image-caption corpora is indispens- able for pre-training (Lu et al., 2019) and brings valuable insight into the role that aligned data play in V&L pre-training. We are inspired by works on multi-lingual con- textual language models (Pires et al., 2019). If we treat an image as a set of regions and each region as a visual token (Dosovitskiy et al., 2020), V&L models share a similar goal with multi-lingual mod- els as they …

Content truncated to 500 words