TimeProVe: Propose-then-Verify Framework for Efficient Long Video Temporal Reasoning in ADL

Forum topic · 小凯 · 2026-06-21

Summary

Long video question answering (LVQA) requires identifying sparse, query-relevant evidence in hours of untrimmed footage. Existing approaches either densely process video with large vision-language models (VLMs), which is computationally expensive, or rely on sparse caption-based reasoning that often misses temporal grounding and motion-centric evidence. TimeProVe is a cost-efficient hybrid framework for long video temporal grounding: a lightweight module first generates action-based candidate answers and evidence hypotheses, and an expensive VLM is then invoked only for targeted verification. At its core, the Action-based Candidate Evidence (ACE) module converts temporally grounded actions into query-conditioned candidate answers and supporting evidence windows via lightweight LLM reasoning. The authors also introduce OpenTSUBench (OTB), an open-ended benchmark for evaluating temporal grounding reasoning in real-world activities of daily living (ADL). Experiments show TimeProVe surpasses the strongest baseline by 7.3% on OTB while reducing VLM calls by 75% and inference cost by 93%. Without explicit temporal grounding training, it achieves competitive performance on Charades-STA and reaches state-of-the-art when combined with a grounding-capable VLM.

Overview

Field: Computer Vision (CV)
Authors: Arkaprava Sinha, Dominick Reilly, Siddharth Krishnan
Published: 2026-06-20
arXiv: 2506.16650

Abstract

Long video question answering (LVQA) requires identifying sparse, query-relevant evidence within hours of untrimmed video. Existing approaches either process video densely with large vision-language models (VLMs), which is computationally prohibitive, or rely on sparse caption-based reasoning, which often misses temporal grounding and motion-centric evidence.

This paper proposes TimeProVe, a cost-efficient hybrid framework for long video temporal grounding reasoning. TimeProVe first uses a lightweight module to generate action-based candidate answers and evidence hypotheses, then invokes an expensive VLM only for targeted verification.

Key Components

ACE (Action-based Candidate Evidence) module: The core of the framework. It converts temporally grounded actions into query-conditioned candidate answers and supporting evidence windows via lightweight LLM reasoning.
OpenTSUBench (OTB): A new open-ended benchmark for evaluating temporal grounding reasoning in real-world activities of daily living (ADL) scenarios.

Results

TimeProVe outperforms the strongest baseline by 7.3% on OTB.
Reduces VLM calls by 75% and inference cost by 93%.
Without explicit temporal grounding training, achieves competitive performance on Charades-STA; combined with a grounding-capable VLM, it reaches state-of-the-art.