AVC-DPO: Aligned Video Captioning via Direct Preference Optimization

TL;DR

AVC-DPO is a post-training framework that aligns video captioning models with human preferences by emphasizing temporal and spatial details, significantly improving detailed captioning performance.

Contribution

The paper introduces AVC-DPO, a novel preference optimization method that enhances video captioning models' focus on human-relevant details through prompt-based training.

Findings

Achieved first place in the Video Detailed Captioning Challenge.

Significantly improved caption alignment with human preferences.

Demonstrated superior performance on the VDC benchmark.

Abstract

Although video multimodal large language models (video MLLMs) have achieved substantial progress in video captioning tasks, it remains challenging to adjust the focal emphasis of video captions according to human preferences. To address this limitation, we propose Aligned Video Captioning via Direct Preference Optimization (AVC-DPO), a post-training framework designed to enhance captioning capabilities in video MLLMs through preference alignment. Our approach designs enhanced prompts that specifically target temporal dynamics and spatial information-two key factors that humans care about when watching a video-thereby incorporating human-centric preferences. AVC-DPO leverages the same foundation model's caption generation responses under varied prompt conditions to conduct preference-aware training and caption alignment. Using this framework, we have achieved exceptional performance in the LOVE@CVPR'25 Workshop Track 1A: Video Detailed Captioning Challenge, achieving first place on the Video Detailed Captioning (VDC) benchmark according to the VDCSCORE evaluation metric.