AI-Generated Video Detection via Perceptual Straightening

2025-07-01

AI Summary

The proliferation of highly realistic videos generated by generative AI poses significant challenges for forensic detection, as existing methods suffer from limited generalization and inadequate modeling of temporal inconsistencies. Method: We propose a novel video authenticity verification paradigm grounded in the geometric properties of neural representations: we observe that authentic videos exhibit more “linearized” temporal trajectories in the representation space of a pretrained Vision Transformer (DINOv2), leading to our “perceptual straightening” hypothesis. We formally characterize this geometric distinction using temporal curvature and step-wise distance metrics—introduced here for the first time—and design a lightweight pipeline that extracts frame-level features, computes statistical geometric descriptors, and feeds them into a compact classifier. Contribution/Results: Our method achieves 97.17% accuracy and 98.63% AUROC on the VidProM benchmark, substantially outperforming state-of-the-art approaches while offering superior generalization, high precision, and low computational overhead.

Technology Category

• Computer Vision: Video Understanding & Activity Analysis
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Web Mining and Content Analysis: Web data provenance, reliability, and authenticity
• Security and Privacy: Data transparency and provenance
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

The rapid advancement of generative AI enables highly realistic synthetic videos, posing significant challenges for content authentication and raising urgent concerns about misuse. Existing detection methods often struggle with generalization and capturing subtle temporal inconsistencies. We propose ReStraV(Representation Straightening Video), a novel approach to distinguish natural from AI-generated videos. Inspired by the "perceptual straightening" hypothesis -- which suggests real-world video trajectories become more straight in neural representation domain -- we analyze deviations from this expected geometric property. Using a pre-trained self-supervised vision transformer (DINOv2), we quantify the temporal curvature and stepwise distance in the model's representation domain. We aggregate statistics of these measures for each video and train a classifier. Our analysis shows that AI-generated videos exhibit significantly different curvature and distance patterns compared to real videos. A lightweight classifier achieves state-of-the-art detection performance (e.g., 97.17% accuracy and 98.63% AUROC on the VidProM benchmark), substantially outperforming existing image- and video-based methods. ReStraV is computationally efficient, it is offering a low-cost and effective detection solution. This work provides new insights into using neural representation geometry for AI-generated video detection.

Problem

• Detecting AI-generated videos via perceptual straightening deviations
• Addressing generalization issues in synthetic video detection methods
• Quantifying temporal inconsistencies using neural representation geometry

Innovation

• Uses perceptual straightening hypothesis for detection
• Analyzes temporal curvature with DINOv2 model
• Lightweight classifier achieves high accuracy