Uncovering Hidden Connections: Iterative Search and Reasoning for Video-grounded Dialog

TL;DR

The paper tackles video-grounded dialog by jointly modeling dialog history and video content to generate accurate answers. It introduces ISR, an architecture with a history-aware textual encoder that uses path search and aggregation, a multimodal iterative visual reasoning network, and a GPT-2–based generator. Key contributions include an interpretable history mining mechanism, a multimodal iterative reasoning module that iteratively refines cross-modal representations, and extensive empirical validation across AVSD DSTC7/8 and VSTAR showing state-of-the-art performance with strong generalization. The approach demonstrates that explicit history pathways and iterative visual reasoning can substantially improve answer quality, with practical implications for robust, interpretable video-grounded dialog systems. The work also provides comprehensive analyses of decoding strategies, input modalities, and efficiency, reinforcing its applicability to real-world multimodal dialogue tasks.

Abstract

In contrast to conventional visual question answering, video-grounded dialog necessitates a profound understanding of both dialog history and video content for accurate response generation. Despite commendable progress made by existing approaches, they still face the challenges of incrementally understanding complex dialog history and assimilating video information. In response to these challenges, we present an iterative search and reasoning framework, which consists of a textual encoder, a visual encoder, and a generator. Specifically, we devise a path search and aggregation strategy in the textual encoder, mining core cues from dialog history that are pivotal to understanding the posed questions. Concurrently, our visual encoder harnesses an iterative reasoning network to extract and emphasize critical visual markers from videos, enhancing the depth of visual comprehension. Finally, we utilize the pre-trained GPT-2 model as our answer generator to decode the mined hidden clues into coherent and contextualized answers. Extensive experiments on three public datasets demonstrate the effectiveness and generalizability of our proposed framework.

Figures (7)

• Figure 1: An illustration of video-grounded dialog.
• Figure 2: Illustration of our proposed framework. It consists of the textual encoder, visual encoder, and generator.
• Figure 3: Illustration of our path search and aggregation strategy.
• Figure 4: Illustration of our multimodal iterative reasoning network.
• Figure 5: Parameter analysis of our model in terms of (a) the number of dialog turns, (b) the selection threshold $p$, (c) the number of iterations $I$.