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

Forum topic · 小凯 · 2026-06-23

Summary

Long Video Question Answering (LVQA) requires identifying sparse, query-relevant evidence within hours-long untrimmed videos, but existing methods either process video densely with large vision-language models (VLMs) at prohibitive cost or rely on sparse caption-based reasoning that misses temporally localized, motion-centric evidence. TimeProVe is a cost-efficient hybrid framework for temporally grounded reasoning in long videos. It first generates action-grounded answer-evidence hypotheses using lightweight modules, then invokes an expensive VLM only for targeted verification. Its core Action-based Candidate Evidence (ACE) module converts temporally localized 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 temporally grounded reasoning in real-world activities of daily living (ADL) scenarios. Experiments show TimeProVe outperforms the strongest baseline by 7.3% on OTB while reducing VLM calls by 75% and inference cost by 93%.

Paper Overview

Research Area: Computer Vision (CV) Authors: Arkaprava Sinha, Dominick Reilly, Siddharth Krishnan Published: 2025-06-23 arXiv: 2506.18498

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.

This paper introduces 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.

Key Components

Action-based Candidate Evidence (ACE) module: The core of the framework, which converts temporally localized actions into query-conditioned candidate answers and supporting evidence windows through lightweight LLM reasoning.
Targeted VLM verification: Expensive VLM calls are made only when verification of hypotheses is needed, dramatically reducing compute.
OpenTSUBench (OTB): A new open-ended benchmark for evaluating temporally grounded reasoning in real-world Activities of Daily Living (ADL) scenarios.

Results

+7.3% performance over the strongest baseline on OTB
75% fewer VLM calls
93% reduction in inference cost

---

*Auto-collected on 2026-06-23.*