What BERT Sees: Cross-Modal Transfer for Visual Question Generation

TL;DR

This paper evaluates BERT's innate ability to perform visual question generation without additional pre-training, demonstrating its adaptability to multi-modal data and its effectiveness in generating questions from visual and textual inputs.

Contribution

The paper introduces BERT-gen, a BERT-based architecture for text generation that works with mono- or multi-modal data without requiring extensive pre-training.

Findings

BERT-gen can effectively generate questions from visual and textual inputs.

The model shows strong performance on VQG datasets with limited data.

BERT's pre-trained knowledge can be leveraged for multi-modal tasks without additional pre-training.

Abstract

Pre-trained language models have recently contributed to significant advances in NLP tasks. Recently, multi-modal versions of BERT have been developed, using heavy pre-training relying on vast corpora of aligned textual and image data, primarily applied to classification tasks such as VQA. In this paper, we are interested in evaluating the visual capabilities of BERT out-of-the-box, by avoiding pre-training made on supplementary data. We choose to study Visual Question Generation, a task of great interest for grounded dialog, that enables to study the impact of each modality (as input can be visual and/or textual). Moreover, the generation aspect of the task requires an adaptation since BERT is primarily designed as an encoder. We introduce BERT-gen, a BERT-based architecture for text generation, able to leverage on either mono- or multi- modal representations. The results reported under different configurations indicate an innate capacity for BERT-gen to adapt to multi-modal data and text generation, even with few data available, avoiding expensive pre-training. The proposed model obtains substantial improvements over the state-of-the-art on two established VQG datasets.