Official implementation of "ReQuest: Rethinking-based Question-Aware Frame Selection for Long-Form Video QA", accepted to ECCV 2026.
ReQuest is a question-aware frame selection framework for long-form video question answering.
The method aims to identify question-relevant visual evidence from long videos and improve video QA performance by selecting informative frames based on rethinking-driven reasoning signals.
The code, pretrained models, and instructions for reproducing the experiments will be released soon.
Please stay tuned.