Publication Date

Spring 2024

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

Sayma Akther

Second Advisor

Fabio Di Troia

Third Advisor

William Andreopoulos

Keywords

Convolutional Neural Networks, Generative Adversarial Networks, Frames Per Second, PSNR, SSIM.

Abstract

Videos are sequences of frames that are displayed continuously within a time frame, which creates the illusion. FPS is defined as the number of frames per second, and is crucial to determine the smoothness of motion or scene changes in the video. To improve the appearance of the videos, we can a technique called Frame Rate Enhancement. This is an approach to augment generated frames between pairs of frames using Generative Adversarial Networks. There are a few traditional techniques using Convolution Neural Networks and Optical Flow based methods, but they create unwanted artifacts such as blurring or ghosting and might reduce the video quality. Hence, we used the ability of GANs to generate high quality intermediate frames that can generate high quality images to increase frame rate. We trained the model on 2 datasets - UCF-101 and Vimeo90K dataset. This enabled the model to learn temporal dependencies and generate realistic frames. Our research proposes a solution to enhance the frame rate in videos with applicability in various scenarios from capturing everyday moments to video surveillance, film production and sports streaming. We finally evaluated the model using 2 metrics - Peak Signal to Noise Ratio and Structural Similarity Index.

Download

Available for download on Sunday, May 25, 2025

Share

COinS