Repository for my graduation work
Full thesis in russian language can be found here - https://drive.google.com/file/d/1dZxL5ZqND9nowRU9bOsN7GGO1fLF3sgb/view?usp=share_link
This work explores different methods of images upscaling, which is poorly solved by classic interpolation but impressive results can be obtained with neural networks.
Main part of my work is working with SRGAN, ESRGAN architectures and training them.
Both models were trained on CelebA dataset. Set5 images and others were used for testing.
Some results of training are shown here.
Most impressive results are visible on this images with ESRGAN. Upscale factor is 4.
