Summary of What Matters in Detecting Ai-generated Videos Like Sora?, by Chirui Chang et al.
What Matters in Detecting AI-Generated Videos like Sora?
by Chirui Chang, Zhengzhe Liu, Xiaoyang Lyu, Xiaojuan Qi
First submitted to arxiv on: 27 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This study investigates the gap between real-world videos and those generated by a state-of-the-art AI model, Stable Video Diffusion. The researchers compare real-world videos with fake ones using three classifiers trained on 3D convolutional networks, each targeting appearance, motion, or geometry. The results show that AI-generated videos are still easily detectable, indicating a significant gap between real and fake videos. Additionally, the study uses Grad-CAM to pinpoint systematic failures of AI-generated videos in appearance, motion, and geometry. An Ensemble-of-Experts model is proposed, integrating information from multiple classifiers for enhanced robustness and generalization ability. The model accurately detects videos generated by Sora without exposure during training, suggesting that the gap can be generalized across various video generative models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This study looks at how well AI-generated videos compare to real ones. They use special computer programs called 3D convolutional networks to look at three things: what the video looks like, how the people and objects move, and how deep they are in the scene. The results show that AI-generated videos aren’t very good yet, because experts can easily tell them apart from real videos. The study also shows where AI-generated videos go wrong, such as not looking quite right or having strange movements. To make AI-generated videos better, the researchers suggest using a combination of these different approaches to detect fake videos. |
Keywords
» Artificial intelligence » Diffusion » Generalization
