Can I Trust Your Answer? Visually Grounded Video Question Answering

Junbin Xiao, Angela Yao, Yicong Li, Tat-Seng Chua; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 13204-13214

We study visually grounded VideoQA in response to the emerging trends of utilizing pretraining techniques for video- language understanding. Specifically by forcing vision- language models (VLMs) to answer questions and simultane- ously provide visual evidence we seek to ascertain the extent to which the predictions of such techniques are genuinely anchored in relevant video content versus spurious corre- lations from language or irrelevant visual context. Towards this we construct NExT-GQA - an extension of NExT-QA with 10.5K temporal grounding (or location) labels tied to the original QA pairs. With NExT-GQA we scrutinize a series of state-of-the-art VLMs. Through post-hoc atten- tion analysis we find that these models are extremely weak in substantiating the answers despite their strong QA per- formance. This exposes the limitation of current VLMs in making reliable predictions. As a remedy we further explore and propose a grounded-QA method via Gaussian mask optimization and cross-modal learning. Experiments with different backbones demonstrate that this grounding mechanism improves both grounding and QA. With these efforts we aim to push towards trustworthy VLMs in VQA systems. Our dataset and code are available at https://github.com/doc-doc/NExT-GQA.