CRAFT: Critic-Refined Adaptive Key-Frame Targeting for Multimodal Video Question Answering

2026-05-18

AI Summary

This work addresses the challenge of locating evidence across heterogeneous videos and providing traceable sources in multi-video question answering grounded in real-world news events. The authors propose CRAFT, a claim-centric framework that integrates query-driven dynamic keyframe selection, a multilingual ASR fallback mechanism, and a hybrid critic loop to iteratively verify and refine claims. The system ultimately outputs a single verified fact along with its complete set of supporting sources through citation-aware fusion. Built as an end-to-end verifiable multimodal QA pipeline, CRAFT incorporates UNLI for temporal entailment, DeBERTa-v3 for cross-claim filtering, and a Llama-3.2-3B adjudicator. It significantly outperforms baselines on both MAGMaR 2026 (overall score: 0.739) and WikiVideo (0.823), demonstrating strong accuracy, fine-grained source attribution, and generalization capability.

Abstract

Grounded multi-video question answering over real-world news events requires systems to surface query-relevant evidence across heterogeneous video archives while attributing every claim to its supporting source. We introduce CRAFT (Critic-Refined Adaptive Key-Frame Targeting), a query-conditioned pipeline that combines dynamic keyframe selection, per-video ASR with multilingual fallback, and a hybrid critic loop to iteratively verify and repair claims before consolidation. The pipeline integrates UNLI temporal entailment, DeBERTa-v3 cross-claim screening, and a Llama-3.2-3B adjudicator, with a final citation-merging stage that emits each fact once with all supporting source identifiers. On MAGMaR 2026, CRAFT achieves the best overall average (0.739), reference recall (0.810), and citation F1 (0.635). We further evaluate on a MAGMaR-style conversion of WikiVideo with 52 non-overlapping event queries, where CRAFT also performs strongly (0.823 Avg), showing that its claim-centric evidence aggregation generalizes beyond MAGMaR. Ablations show that atomic claims, ASR, and the critic loop drive the main gains over the vanilla query-conditioned baseline. Code and implementation details are publicly available at https://github.com/bhosalems/CRAFT.

Problem

• multimodal video question answering
• evidence retrieval
• source attribution
• grounded reasoning
• heterogeneous video archives

Innovation

• adaptive key-frame targeting
• critic-refined verification
• multimodal video QA
• claim-centric evidence aggregation
• multilingual ASR fallback