RealAIGI: Realistic AI Generated Image Dataset
- Citation Author(s):
-
Md Mahinur Alam
(Kumoh National Institute of Technology, Gumi, South Korea)
Taesoo Jun
(Kumoh National Institute of Technology, Gumi, South Korea)
- Submitted by:
- Taesoo Jun
- Last updated:
- DOI:
- 10.21227/0da4-g645
23 views
- Categories:
- Keywords:
Abstract
The rapid advancement of generative neural networks has facilitated the creation of photorealistic images, raising concerns about the proliferation of misinformation. Detecting AI-generated fakes has become crucial, given their potential impact on public opinion and various sectors. This dataset presents a comparative analysis of real and AI-generated images, focusing on building a novel dataset named Realistic AI-Generated Image (RealAIGI) dataset. Leveraging sophisticated diffusion transformer models like Sora, the dataset comprises diverse general content, aiming to match the scale of existing general image datasets. We discuss the significance of such datasets in enhancing detection systems' robustness and real-world applicability, this work contributes to advancing AI-generated image detection.
Instructions:
This dataset is specifically curated for the development and evaluation of machine learning models aimed at detecting AI-generated images. It comprises two distinct categories of images: AI-generated and authentic real-world images. Images are organized into various classes, enabling detailed analysis and classification.
Dataset Structure
The dataset is divided into three subsets:
- Training Set: Used for model training and parameter tuning.
- Validation Set: Employed for validating the model's performance during training.
- Test Set: Reserved for the final unbiased evaluation of model performance.
Each subset contains two primary categories:
- AI-generated images
- Real images
