Final project for Biomedical Imaging & Analytics and Deep Learning courses (Skoltech, 2022).
Science in medicine is evolving rapidly, not only in the direction of new approaches or algorithms, but also in the direction of medical devices. One of the hot issues that have been raised is the problem of domain adaptation. Algorithms trained on old devices do not show good results on new ones, due to differences between the domains. The solution to this problem in the minds of scientists is not only in the medical field, but most approaches are not applicable in this area. This paper shows, that the style can be transferred using Image-to-Image translation and Transfer Learning. The aim of the project is to review current methods for Image-to-Image Translation and compare its results with Transfer Learning. As a result we show that Image-to-Image Translation is better for domain adaption than Transfer Learning with small amount of labeled data.
- Vladimir Baikalov
- Suraj Singh
- Maksim Pakhomov
- Vladimir Shaposhnikov
- Gleb Bazhenov
- Dmitrii Gavrilev
- Saydash Miftakhov