| Form of studies |
Bachelor |
| Title of the study programm |
Information Technology |
| Title in original language |
Ģeneratīvo konkurējošo tīklu analīze un salīdzinājums video satura ģenerēšanā |
| Title in English |
Analysis and Comparison of Generative Adversarial Networks for the Video Content Generation |
| Department |
Faculty Of Computer Science Information Tehnology And Energy |
| Scientific advisor |
Sintija Petroviča-Kļaviņa |
| Reviewer |
Oļesja Večerinska |
| Abstract |
This bachelor thesis explores the application of Generative Adversarial Networks
(GANs) in video content generation. In recent years, this technology has significantly
influenced the development of artificial intelligence, particularly in the creation of
synthetic multimedia content. Various GAN architectures are analyzed in the context
of video generation, including their advantages and limitations. The core issue
addressed in this research is the challenge of generating high-quality, realistic motion,
as well as comparing algorithms based on different performance metrics. The aim of
the thesis is to develop a comparison methodology and experimentally evaluate the
suitability of different GAN models for specific video generation tasks. The work
includes a literature review, the design of a comparison approach, and practical
experiments using StyleGAN3, DCGAN, and VGAN models. The experimental part
covers video data generation, evaluation using FID and IS metrics, and analysis of the
obtained results. The conclusion summarizes the most suitable architectures and
identifies potential directions for future improvements in video generation.
The bachelor's thesis consists of 58 pages, 22 figures, 2 tables, 1 appendix, and
42 references. |
| Keywords |
Ģeneratīvie konkurējošie tīkli, video satura ģenerēšana, mākslīgais intelekts, StyleGAN, DCGAN, VGAN. |
| Keywords in English |
Generative Adversarial Networks, video content generation, artificial intelligence, StyleGAN, DCGAN, VGAN. |
| Language |
lv |
| Year |
2025 |
| Date and time of uploading |
22.05.2025 10:41:13 |
