Open Access
American Research Journal of Computer Science and Information Technology
ISSN (Online): 2572-2921
DOI: 10.46568/arjcsit
Pretrained Diffusion Models for Image Segmentation
Lead ML Engineer, Ticket to the moon, Inc, Moscow, Russia.
Gainetdionov Ainur Fanurovich, “Pretrained Diffusion Models for Image Segmentation”, American Research
Journal of Computer Science and Information Technology, Vol 7, no. 1, 2024, pp. 63-67.
Abstract
This article discusses the use of pre-trained diffusion models for image segmentation. The study aims to increase the
accuracy of segmentation and reduce required training data by using diffusion methods. During the work, models based
on noise generation and removal were used, which allows for improvement in the quality of segmentation masks. The
methodology includes the use of U-Net and Transformer algorithms, which contribute to the creation of high-precision
segmentation masks of objects in images. The results showed that the use of pre-prepared models significantly improved
the accuracy and consistency of segmentation masks compared to traditional methods. In conclusion, it is noted that the
proposed approach provides higher segmentation efficiency without requiring extensive data, which opens up prospects
for further technology development in various fields..
