Retrieval-Driven Training-Free AI-Generated Video Attribution
Quick Answer
This paper presents a training-free paradigm for attributing AI-generated videos, treating it as an instance retrieval task.
Quick Take
The proposed method achieves a Rank-1 accuracy of 20.5% and a mean Average Precision of 16.6% on the GenVidBench benchmark, outperforming existing state-of-the-art techniques.
Key Points
- Introduces a generative fingerprint-based pipeline for video attribution.
- Utilizes orthogonal color transformation and multi-scale quantized residual generation.
- Achieves superior performance compared to existing methods in video detection.
- Focuses on addressing cybersecurity threats from realistic AI-generated videos.
- Code available for public access to facilitate further research.
DeepSignal Analysis
What happened
The paper introduces a training-free method for attributing AI-generated videos, framing it as an instance retrieval task. The proposed approach achieves a Rank-1 accuracy of 20.5% and a mean Average Precision of 16.6% on the GenVidBench benchmark, surpassing existing techniques.
Key evidence
- The method focuses on AI-generated videos, which are increasingly realistic and pose threats to cybersecurity and social governance.
- The proposed pipeline includes an adapted orthogonal color transformation and temporal-semantic aggregation to capture artifacts from generative models.
- Experimental results show that the new method outperforms state-of-the-art techniques, achieving a Rank-1 accuracy of 20.5% and a mean Average Precision of 16.6%.
Why it matters
As AI-generated videos become more realistic, the ability to attribute them to their sources is crucial for forensic investigations and legal regulations. This method addresses limitations of existing attribution techniques that primarily focus on images, thereby enhancing the ability to manage the risks associated with video misuse.
What to watch
Paper Resources
Source Excerpt
AI-generated videos are becoming increasingly realistic and difficult to distinguish from authentic ones, which facilitates malicious misuse and poses growing threats to cybersecurity and social governance. Attributing AI-generated videos to their specific generative sources is therefore of critical importance for forensic investigation and legal regulation. However, most existing visual attribution methods focus on images and particularly rely on the image generation model, thereby lacking the
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.