OmniVL-Guard: Towards Unified Vision-Language Forgery Detection and Grounding via Balanced RL

TL;DR

OmniVL-Guard introduces a balanced reinforcement learning framework that enhances unified vision-language forgery detection and grounding, effectively addressing the challenge of multi-task optimization in multimodal misinformation detection.

Contribution

It proposes a novel balanced RL framework with Self-Evolving CoT Generation and ARSPO to improve multi-task learning in vision-language forgery detection and grounding.

Findings

Outperforms state-of-the-art methods in experiments.

Demonstrates zero-shot robustness across out-of-domain scenarios.

Achieves significant improvements in fine-grained grounding accuracy.

Abstract

Existing forgery detection methods are often limited to uni-modal or bi-modal settings, failing to handle the interleaved text, images, and videos prevalent in real-world misinformation. To bridge this gap, this paper targets to develop a unified framework for omnibus vision-language forgery detection and grounding. In this unified setting, the {interplay} between diverse modalities and the dual requirements of simultaneous detection and localization pose a critical ``difficulty bias`` problem: the simpler veracity classification task tends to dominate the gradients, leading to suboptimal performance in fine-grained grounding during multi-task optimization. To address this challenge, we propose \textbf{OmniVL-Guard}, a balanced reinforcement learning framework for omnibus vision-language forgery detection and grounding. Particularly, OmniVL-Guard comprises two core designs: Self-Evolving CoT Generatio and Adaptive Reward Scaling Policy Optimization (ARSPO). {Self-Evolving CoT Generation} synthesizes high-quality reasoning paths, effectively overcoming the cold-start challenge. Building upon this, {Adaptive Reward Scaling Policy Optimization (ARSPO)} dynamically modulates reward scales and task weights, ensuring a balanced joint optimization. Extensive experiments demonstrate that OmniVL-Guard significantly outperforms state-of-the-art methods and exhibits zero-shot robust generalization across out-of-domain scenarios. The dataset and code are publicly available at https://github.com/shen8424/OmniVL-Guard.