Abstract: The quest for high-quality video content necessitates the advancement of techniques to improve both spatial and temporal resolutions of videos. Traditional methods often falter in enhancing detail and fluidity simultaneously in low-resolution and low-frame-rate videos. This Master's thesis will used an innovative approach to Space-Time Video Super-Resolution (STVSR) employing modern artificial intelligence techniques, notably deformable convolution, focusing on neighborhood frames. The technique utilizes deformable convolutional networks (DCNs) to precisely align features across frames, enabling accurate prediction of intermediate frames and improved reconstruction of high-resolution content.

Supervisor: Adam Novozámský