Year: 2025
Type: dissertation
Abstract: The paper research the creation of a detector method that allows to distinguish between real images and images generated by diffusion models. The novelty of the research is the creation of new architectures capable of effectively identifying and classifying features characteristic of synthetic images, as well as exploring the possibility of using a textual description to improve accuracy. The aim of the work is to study architectures and algorithms that allow detecting images generated by diffusion models. To achieve the research goal, the following tasks were performed: a review of modern methods; an architecture and algorithm were created; a dataset was selected and prepared; experiments were carried out; experimental results and metrics were analyzed.
Authors: Горбачев, А. Н., Gorbachev, A. N.
Language: Russian
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7117191470