Salvato in:
Dettagli Bibliografici
Autori principali: Jain, Jitesh, Li, Jialuo, Ma, Zixian, Zhang, Jieyu, Kim, Chris Dongjoo, Lee, Sangho, Tripathi, Rohun, Gupta, Tanmay, Clark, Christopher, Shi, Humphrey
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2512.13874
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918416058679296
author Jain, Jitesh
Li, Jialuo
Ma, Zixian
Zhang, Jieyu
Kim, Chris Dongjoo
Lee, Sangho
Tripathi, Rohun
Gupta, Tanmay
Clark, Christopher
Shi, Humphrey
author_facet Jain, Jitesh
Li, Jialuo
Ma, Zixian
Zhang, Jieyu
Kim, Chris Dongjoo
Lee, Sangho
Tripathi, Rohun
Gupta, Tanmay
Clark, Christopher
Shi, Humphrey
contents As humans, we are natural any-horizon reasoners, i.e., we can decide whether to iteratively skim long videos or watch short ones in full when necessary for a given task. With this in mind, one would expect video reasoning models to reason flexibly across different durations. However, SOTA models are still trained to predict answers in a single turn while processing a large number of frames, akin to watching an entire long video, requiring significant resources. This raises the question: Is it possible to develop performant any-horizon video reasoning systems? Inspired by human behavior, we first propose SAGE, an agent system that performs multi-turn reasoning on long videos while handling simpler problems in a single turn. Secondly, we introduce an easy synthetic data generation pipeline using Gemini-2.5-Flash to train the orchestrator, SAGE-MM, which lies at the core of SAGE. We further propose an effective RL post-training recipe essential for instilling any-horizon reasoning ability in SAGE-MM. Thirdly, we curate SAGE-Bench with an average duration of greater than 700 seconds for evaluating video reasoning ability in real-world entertainment use cases. Lastly, we empirically validate the effectiveness of our system, data, and RL recipe, observing notable improvements of up to 6.1% on open-ended video reasoning tasks, as well as an impressive 8.2% improvement on videos longer than 10 minutes.
format Preprint
id arxiv_https___arxiv_org_abs_2512_13874
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SAGE: Training Smart Any-Horizon Agents for Long Video Reasoning with Reinforcement Learning
Jain, Jitesh
Li, Jialuo
Ma, Zixian
Zhang, Jieyu
Kim, Chris Dongjoo
Lee, Sangho
Tripathi, Rohun
Gupta, Tanmay
Clark, Christopher
Shi, Humphrey
Computer Vision and Pattern Recognition
As humans, we are natural any-horizon reasoners, i.e., we can decide whether to iteratively skim long videos or watch short ones in full when necessary for a given task. With this in mind, one would expect video reasoning models to reason flexibly across different durations. However, SOTA models are still trained to predict answers in a single turn while processing a large number of frames, akin to watching an entire long video, requiring significant resources. This raises the question: Is it possible to develop performant any-horizon video reasoning systems? Inspired by human behavior, we first propose SAGE, an agent system that performs multi-turn reasoning on long videos while handling simpler problems in a single turn. Secondly, we introduce an easy synthetic data generation pipeline using Gemini-2.5-Flash to train the orchestrator, SAGE-MM, which lies at the core of SAGE. We further propose an effective RL post-training recipe essential for instilling any-horizon reasoning ability in SAGE-MM. Thirdly, we curate SAGE-Bench with an average duration of greater than 700 seconds for evaluating video reasoning ability in real-world entertainment use cases. Lastly, we empirically validate the effectiveness of our system, data, and RL recipe, observing notable improvements of up to 6.1% on open-ended video reasoning tasks, as well as an impressive 8.2% improvement on videos longer than 10 minutes.
title SAGE: Training Smart Any-Horizon Agents for Long Video Reasoning with Reinforcement Learning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.13874