DBINDS——扩散模型反演的初始噪声能否帮助识别AI生成视频？
DBINDS -- Can Initial Noise from Diffusion Model Inversion Help Reveal AI-Generated Videos?
AI总结：
针对AI生成视频的取证挑战，提出基于扩散模型反演的DBINDS检测器，通过分析真实与生成视频初始噪声序列的系统性差异，构建INDS并提取多域多尺度特征，结合优化分类器实现跨生成器的强泛化性能。
摘要：
AI生成视频发展迅速，对内容安全和取证分析构成严峻挑战。现有检测器主要依赖像素级视觉线索，对未见过的生成器泛化能力差。我们提出DBINDS，一种基于扩散模型反演的检测器，分析潜在空间动态而非像素。我们发现扩散反演恢复的初始噪声序列在真实与生成视频间存在系统性差异。基于此，DBINDS构建初始噪声差异序列（INDS）并提取多域、多尺度特征。通过特征优化和贝叶斯搜索调优的LightGBM分类器，DBINDS（在单一生成器上训练）在GenVidBench上实现了强大的跨生成器性能，在有限数据场景中展现出良好的泛化性和鲁棒性。
英文摘要：
AI-generated video has advanced rapidly and poses serious challenges to content security and forensic analysis. Existing detectors rely mainly on pixel-level visual cues and generalize poorly to unseen generators. We propose DBINDS, a diffusion-model-inversion based detector that analyzes latent-space dynamics rather than pixels. We find that initial noise sequences recovered by diffusion inversion differ systematically between real and generated videos. Building on this, DBINDS forms an Initial Noise Difference Sequence (INDS) and extracts multi-domain, multi-scale features. With feature optimization and a LightGBM classifier tuned by Bayesian search, DBINDS (trained on a single generator) achieves strong cross-generator performance on GenVidBench, demonstrating good generalization and robustness in limited-data settings.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
