Summary of Tackling Structural Hallucination in Image Translation with Local Diffusion, by Seunghoi Kim et al.
Tackling Structural Hallucination in Image Translation with Local Diffusion
by Seunghoi Kim, Chen Jin, Tom Diethe, Matteo Figini, Henry F. J. Tregidgo, Asher Mullokandov, Philip Teare, Daniel C. Alexander
First submitted to arxiv on: 9 Apr 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 Medium Difficulty summary: Recent advancements in diffusion models have significantly improved conditioned image generation; however, they struggle to reconstruct out-of-distribution (OOD) images, such as unseen tumors in medical images, which can lead to “image hallucination” and misdiagnosis. We propose a training-free diffusion framework that reduces hallucination by using multiple Local Diffusion processes, involving OOD estimation followed by a “branching” module for generating local predictions and a “fusion” module for integrating them into one. Our evaluation demonstrates the effectiveness of our method in mitigating hallucinations over baseline models quantitatively and qualitatively, reducing misdiagnosis rates by 40% and 25% in medical and natural image datasets, respectively. This approach is compatible with various pre-trained diffusion models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: Imagine trying to generate a new image of something you’ve never seen before, like a tumor in an X-ray. Computers are getting better at doing this, but sometimes they make mistakes by creating fake images instead. We want to fix this problem so doctors can get accurate diagnoses. To do this, we developed a new way of using computer models that generates multiple possible images and then combines them into one realistic image. This helps reduce the number of wrong predictions made by these computers, which is important for medical imaging. |
Keywords
» Artificial intelligence » Diffusion » Hallucination » Image generation
