SynPO: Synergizing Descriptiveness and Preference Optimization for Video Detailed Captioning

TL;DR

This paper introduces SynPO, a new preference optimization method that improves fine-grained video captioning by balancing detailed description generation and training efficiency, outperforming existing methods.

Contribution

We propose SynPO, a novel optimization approach that enhances video captioning models by preventing negative preferences from dominating and maintaining language capabilities, with improved training efficiency.

Findings

SynPO outperforms DPO variants on video captioning benchmarks.

SynPO achieves 20% higher training efficiency.

SynPO generalizes well across NLP tasks and models.

Abstract

Fine-grained video captioning aims to generate detailed, temporally coherent descriptions of video content. However, existing methods struggle to capture subtle video dynamics and rich detailed information. In this paper, we leverage preference learning to enhance the performance of vision-language models in fine-grained video captioning, while mitigating several limitations inherent to direct preference optimization (DPO). First, we propose a pipeline for constructing preference pairs that leverages the intrinsic properties of VLMs along with partial assistance from large language models, achieving an optimal balance between cost and data quality. Second, we propose Synergistic Preference Optimization (SynPO), a novel optimization method offering significant advantages over DPO and its variants. SynPO prevents negative preferences from dominating the optimization, explicitly preserves the model's language capability to avoid deviation of the optimization objective, and improves training efficiency by eliminating the need for the reference model. We extensively evaluate SynPO not only on video captioning benchmarks (e.g., VDC, VDD, VATEX) but also across well-established NLP tasks, including general language understanding and preference evaluation, using diverse pretrained models. Results demonstrate that SynPO consistently outperforms DPO variants while achieving 20\% improvement in training efficiency. Code is available at https://github.com/longmalongma/SynPO