TimeProVe: Propose, then Verify for Efficient Long Video Temporal Reasoning in Activities of Daily Living

2026-06-18

AI Summary

This work addresses the challenge of long-form video question answering, which requires locating sparse, query-relevant evidence within hours of unedited video. Existing approaches are either computationally expensive or rely on sparse subtitles, thereby losing critical temporal and action-related information. To overcome these limitations, the authors propose TimeProVe, a novel “propose-then-verify” hybrid framework that first generates lightweight answer-evidence hypotheses via an Action-driven Candidate Evidence (ACE) module and then selectively invokes a large language model for targeted verification—enabling efficient and accurate reasoning without explicit temporal annotations. Evaluated on the newly introduced open-domain ADL benchmark OpenTSUBench, TimeProVe outperforms the strongest baseline by 7.3% in accuracy while reducing large model invocations by 75% and overall inference cost by 93%. It also achieves state-of-the-art performance on Charades-STA.

Abstract

Long Video Question Answering (LVQA) requires identifying sparse, query-relevant evidence within hours-long untrimmed videos. Existing approaches either process videos densely with large vision-language models (VLMs), incurring prohibitive computational cost, or rely on sparse caption-based reasoning, which often misses temporally localized and motion-centric evidence. We introduce TimeProVe, a cost-efficient hybrid framework for temporally grounded reasoning in long videos. TimeProVe first employs lightweight modules to generate action-grounded answer--evidence hypotheses and subsequently invokes an expensive VLM only for targeted verification. The core of our framework lies in the Action-based Candidate Evidence (ACE) module, which converts temporally localized actions into query-conditioned candidate answers and supporting evidence windows through lightweight LLM reasoning. We further introduce OpenTSUBench (OTB), an open-ended benchmark designed to evaluate temporally grounded reasoning in real-world Activities of Daily Living (ADL) scenarios. Experiments show that TimeProVe outperforms the strongest baseline on OTB by 7.3%, while reducing VLM calls by 75% and inference cost by 93%. Furthermore, without explicit temporal grounding training, TimeProVe achieves competitive performance on Charades-STA, and reaches state-of-the-art results when enhanced with grounding VLMs.

Problem

• Long Video Question Answering
• Temporal Reasoning
• Activities of Daily Living
• Time Localization
• Video Understanding

Innovation

• TimeProVe
• temporal reasoning
• long video question answering
• hybrid verification framework
• action-grounded evidence