Can I Trust Your Answer? Visually Grounded Video Question Answering
This addresses the trustworthiness of VLMs in VideoQA systems, exposing limitations in current models and providing a dataset and method to enhance reliability, though it is incremental in improving existing techniques.
The authors tackled the problem of verifying whether vision-language models (VLMs) for video question answering (VideoQA) rely on relevant visual evidence or spurious correlations, by constructing NExT-GQA with 10.5K temporal grounding labels and analyzing state-of-the-art models. They found these models are extremely weak in substantiating answers despite strong QA performance, and proposed a grounded-QA method via Gaussian mask optimization and cross-modal learning that improves both grounding and QA.
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 simultaneously 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 correlations from language or irrelevant visual context. Towards this, we construct NExT-GQA -- an extension of NExT-QA with 10.5$K$ 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 attention analysis, we find that these models are extremely weak in substantiating the answers despite their strong QA performance. 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.