Towards a Unified Model for Generating Answers and Explanations in Visual Question Answering
This addresses the issue of inconsistency and lack of grounding in VQA systems for researchers and practitioners, though it is incremental as it builds on existing multitask and fine-tuning approaches.
The paper tackles the problem of generating both answers and explanations in visual question answering by proposing a unified multitask learning model, which improves answer accuracy on A-OKVQA by 10-15% and achieves state-of-the-art explanation scores on multiple datasets.
The field of visual question answering (VQA) has recently seen a surge in research focused on providing explanations for predicted answers. However, current systems mostly rely on separate models to predict answers and generate explanations, leading to less grounded and frequently inconsistent results. To address this, we propose a multitask learning approach towards a Unified Model for Answer and Explanation generation (UMAE). Our approach involves the addition of artificial prompt tokens to training data and fine-tuning a multimodal encoder-decoder model on a variety of VQA-related tasks. In our experiments, UMAE models surpass the prior state-of-the-art answer accuracy on A-OKVQA by 10~15%, show competitive results on OK-VQA, achieve new state-of-the-art explanation scores on A-OKVQA and VCR, and demonstrate promising out-of-domain performance on VQA-X.