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

Source: [CRAFT: Critic-Refined Adaptive Key-Frame Targeting for Multimodal Video Question Answering](<https://www.sota2.com/research/paper/6a0d3feff984739ed6737669>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-21T05:00:05.413Z

Authors: Mahesh Bhosale, Abdul Wasi, Vishvesh Trivedi, Pengyu Yan, Akhil Gorugantu, David Doermann

Published: 2026-05-18T20:01:05.000Z

## 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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Grounded Multi-Video Question Answering on MAGMaR (test)](<https://www.sota2.com/research/sota/grounded-multi-video-question-answering-on-magmar-test.md>): Grounded Multi-Video Question Answering; MAGMaR (test); Reference Precision (Ref-P): 82.2.

- [Grounded Multi-Video Question Answering on WikiVideo](<https://www.sota2.com/research/sota/grounded-multi-video-question-answering-on-wikivideo.md>): Grounded Multi-Video Question Answering; WikiVideo; Ref Precision: 94.

- [Report Generation on MAGMaR (test)](<https://www.sota2.com/research/sota/report-generation-on-magmar-test.md>): Report Generation; MAGMaR (test); ROUGE-L: 18.39.

- [Report Generation on WikiVideo](<https://www.sota2.com/research/sota/report-generation-on-wikivideo.md>): Report Generation; WikiVideo; ROUGE-L: 30.14.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2605.19075>)

[PDF](<https://arxiv.org/pdf/2605.19075.pdf>)