🤖 AI Summary
Video dialogue understanding faces the challenge of jointly modeling conversational history and video content. To address this, we propose an iterative search-and-reasoning framework with three key innovations: (1) a novel path-search-and-aggregation text encoding mechanism that explicitly identifies salient dialogue cues; (2) an iterative visual encoder that dynamically enhances key frames and salient regions through multi-step refinement; and (3) implicit cross-modal alignment and progressive fusion of textual and visual clues. The framework integrates a specialized text encoder, the iterative visual encoder, a GPT-2-based generator, and a multi-stage alignment module. Evaluated on three public video dialogue benchmarks, our method achieves absolute improvements of 3.2–5.7 percentage points in accuracy and BLEU score over state-of-the-art approaches, demonstrating both effectiveness and strong generalization capability.
📝 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.
