Summary of Looks Too Good to Be True: An Information-theoretic Analysis Of Hallucinations in Generative Restoration Models, by Regev Cohen et al.
Looks Too Good To Be True: An Information-Theoretic Analysis of Hallucinations in Generative Restoration Models
by Regev Cohen, Idan Kligvasser, Ehud Rivlin, Daniel Freedman
First submitted to arxiv on: 26 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
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 The paper investigates the phenomenon of hallucinations in generative models that produce visually indistinguishable results from real data. These models are capable of producing high-perceptual-quality images, but they also exhibit a growing tendency to generate realistic-looking details that do not exist in the ground truth images. The authors rigorously analyze the relationship between uncertainty and perception, revealing a fundamental tradeoff between the two factors. They define the inherent uncertainty of the restoration problem and show that attaining perfect perceptual quality entails at least twice this uncertainty. Additionally, they establish a relation between distortion, uncertainty, and perception, proving that the uncertainly-perception tradeoff induces the well-known perception-distortion tradeoff. Experiments with super-resolution and inpainting algorithms demonstrate the theoretical findings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how computers can make pictures look real, but sometimes these computers get confused and add fake details to the picture. The researchers looked at why this happens and found that there’s a trade-off between making the picture look good and keeping it accurate. They showed that if you want the picture to be perfect, you need to sacrifice some accuracy. This means that people using computer algorithms for tasks like super-resolution or inpainting should know about these limitations so they can make informed decisions. |
Keywords
» Artificial intelligence » Super resolution
