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

2026-02-02

AI Summary

This work addresses the challenge of limited long-video annotations, which hinders short-context vision-language models from comprehending ultra-long videos. To overcome this, the authors propose a two-stage unsupervised preference optimization framework. In the first stage, key short video clips are anchored to generate and filter synthetic preference triplets. The second stage leverages recursive captioning and large language models to construct multi-segment reasoning tasks, enabling self-supervised preference alignment over long videos. Requiring no human annotations and using only 16K synthetic samples, the method outperforms current open-source state-of-the-art approaches on multiple long-video benchmarks while maintaining strong performance on short-video tasks such as MVBench, significantly enhancing semantic understanding of ultra-long videos.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Preferences

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

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.

Problem

• long-form video understanding
• preference optimization
• vision-language models
• annotation-free learning
• video reasoning

Innovation

• Direct Preference Optimization
• Long-Form Video Understanding
• Synthetic Preference Data
• Recursive Captioning
• Multi-Segment Reasoning