SPLIT: Training-Free AI-Generated and Partially Edited Video Detection via Spatial Patch-Level Incoherence and Temporal Roughness
📰 ArXiv cs.AI
Learn to detect AI-generated and partially edited videos using SPLIT, a training-free method that analyzes spatial and temporal inconsistencies
Action Steps
- Extract patch tokens from a frozen vision encoder to analyze spatial inconsistencies
- Compute spatial patch-level incoherence to identify potential AI-generated regions
- Analyze temporal roughness to detect edited or generated video segments
- Combine spatial and temporal features to make a detection decision
- Evaluate the performance of SPLIT on a dataset of real and AI-generated videos
Who Needs to Know This
Video analysts, AI researchers, and developers working on video authentication and forensics can benefit from this method to improve the accuracy of their systems
Key Insight
💡 SPLIT detects AI-generated and partially edited videos by analyzing spatial patch-level incoherence and temporal roughness, without requiring training data
Share This
📹 Detect AI-generated videos with SPLIT, a training-free method that analyzes spatial and temporal inconsistencies! #AI #VideoAuthentication
Key Takeaways
Learn to detect AI-generated and partially edited videos using SPLIT, a training-free method that analyzes spatial and temporal inconsistencies
Full Article
Title: SPLIT: Training-Free AI-Generated and Partially Edited Video Detection via Spatial Patch-Level Incoherence and Temporal Roughness
Abstract:
arXiv:2607.02886v1 Announce Type: cross Abstract: Deploying AI-generated video detectors in real-world services demands an ultra-low false positive rate (FPR) on real videos to avoid falsely rejecting authentic content, a regime where standard metrics such as AUROC fail to reflect actual operating behavior. We introduce Spatial Patch-Level Incoherence and Temporal Roughness (SPLIT), a training-free detector that operates on patch tokens from a frozen vision encoder to detect both fully generated
Abstract:
arXiv:2607.02886v1 Announce Type: cross Abstract: Deploying AI-generated video detectors in real-world services demands an ultra-low false positive rate (FPR) on real videos to avoid falsely rejecting authentic content, a regime where standard metrics such as AUROC fail to reflect actual operating behavior. We introduce Spatial Patch-Level Incoherence and Temporal Roughness (SPLIT), a training-free detector that operates on patch tokens from a frozen vision encoder to detect both fully generated
DeepCamp AI