Summary of Imaginet: a Multi-content Benchmark For Synthetic Image Detection, by Delyan Boychev et al.
ImagiNet: A Multi-Content Benchmark for Synthetic Image Detection
by Delyan Boychev, Radostin Cholakov
First submitted to arxiv on: 29 Jul 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: 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 A recent surge in generative models has made it challenging to distinguish between real and synthetic images, leading to a pressing need for robust detectors. Current datasets used to train these detectors are often limited by low-quality generated images or a focus on a single content type, resulting in poor generalizability. The authors introduce ImagiNet, a comprehensive dataset of 200K examples spanning four categories: photos, paintings, faces, and miscellaneous. Synthetic images were created using both open-source and proprietary generators, while real counterparts were sourced from public datasets. ImagiNet allows for two-track evaluation: classification as real or synthetic, and identification of the generative model. To establish a strong baseline, the authors trained a ResNet-50 model using a self-supervised contrastive objective (SelfCon) for each track, achieving AUCs up to 0.99 and balanced accuracy ranging from 86% to 95%. The provided model shows generalizability, achieving zero-shot state-of-the-art performance on previous synthetic detection benchmarks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Generative models can create images that look very real. This is both amazing and a little scary because it means we need ways to tell what’s real and what’s not. Right now, the tools we use to do this are limited because they’re trained on low-quality fake images or just one type of image. That makes them not very good at recognizing other types of fake images. The authors of this paper created a new dataset with 200,000 examples of different types of images, including photos, paintings, faces, and more. They used both free and paid tools to create the fake images, and they took real images from public sources. This dataset lets us evaluate these image recognition tools in two ways: can we tell if it’s a real or fake image, and can we figure out what tool was used to create the fake one? The authors also showed that their approach works well on other datasets by achieving high scores without needing any extra training. |
Keywords
» Artificial intelligence » Classification » Generative model » Resnet » Self supervised » Zero shot
