LongVPO: From Anchored Cues to Self-Reasoning for Long-Form Video Preference Optimization

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Huang, Zhenpeng, Li, Jiaqi, Jia, Zihan, Li, Xinhao, Meng, Desen, Song, Lingxue, Chen, Xi, Li, Liang, Wang, Limin
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915768924372992
author Huang, Zhenpeng
Li, Jiaqi
Jia, Zihan
Li, Xinhao
Meng, Desen
Song, Lingxue
Chen, Xi
Li, Liang
Wang, Limin
author_facet Huang, Zhenpeng
Li, Jiaqi
Jia, Zihan
Li, Xinhao
Meng, Desen
Song, Lingxue
Chen, Xi
Li, Liang
Wang, Limin
contents We present LongVPO, a novel two-stage Direct Preference Optimization framework that enables short-context vision-language models to robustly understand ultra-long videos without any long-video annotations. In Stage 1, we synthesize preference triples by anchoring questions to individual short clips, interleaving them with distractors, and applying visual-similarity and question-specificity filtering to mitigate positional bias and ensure unambiguous supervision. We also approximate the reference model's scoring over long contexts by evaluating only the anchor clip, reducing computational overhead. In Stage 2, we employ a recursive captioning pipeline on long videos to generate scene-level metadata, then use a large language model to craft multi-segment reasoning queries and dispreferred responses, aligning the model's preferences through multi-segment reasoning tasks. With only 16K synthetic examples and no costly human labels, LongVPO outperforms the state-of-the-art open-source models on multiple long-video benchmarks, while maintaining strong short-video performance (e.g., on MVBench), offering a scalable paradigm for efficient long-form video understanding.
format Preprint
id arxiv_https___arxiv_org_abs_2602_02341
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle LongVPO: From Anchored Cues to Self-Reasoning for Long-Form Video Preference Optimization
Huang, Zhenpeng
Li, Jiaqi
Jia, Zihan
Li, Xinhao
Meng, Desen
Song, Lingxue
Chen, Xi
Li, Liang
Wang, Limin
Computer Vision and Pattern Recognition
We present LongVPO, a novel two-stage Direct Preference Optimization framework that enables short-context vision-language models to robustly understand ultra-long videos without any long-video annotations. In Stage 1, we synthesize preference triples by anchoring questions to individual short clips, interleaving them with distractors, and applying visual-similarity and question-specificity filtering to mitigate positional bias and ensure unambiguous supervision. We also approximate the reference model's scoring over long contexts by evaluating only the anchor clip, reducing computational overhead. In Stage 2, we employ a recursive captioning pipeline on long videos to generate scene-level metadata, then use a large language model to craft multi-segment reasoning queries and dispreferred responses, aligning the model's preferences through multi-segment reasoning tasks. With only 16K synthetic examples and no costly human labels, LongVPO outperforms the state-of-the-art open-source models on multiple long-video benchmarks, while maintaining strong short-video performance (e.g., on MVBench), offering a scalable paradigm for efficient long-form video understanding.
title LongVPO: From Anchored Cues to Self-Reasoning for Long-Form Video Preference Optimization
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2602.02341