Temporally consistent FastDVDNet: an overlap loss implementation for FastDVDNet

<p>The objective of this thesis is the improvement of FastDVDNet’s temporal performance within the video denoising problem space. Video denoising refers to the removal of undesired artifacts, or noise, from a given video sequence. Due to the temporal nature of video sequences, flickering or temporal artifacts remain after denoising. TempFormer, another video denoising model, recently created a solution that provides state of the art results for the minimization of flickering. After modification of FastDVDNet, the goal is a model with fast inference time, or time to denoise a frame, and reduced flickering. The solution will primarily cover the modification of FastDVDNet’s high-level architecture and loss function. Additionally, the problem space of video denoising and methods of determining a model’s performance are explored. The content will focus on a grounds-up approach for both understanding and reproducing the results that follow.</p>

Name thesis:degree_name

Master of Science (MS) in Computer Science

Level thesis:degree_level

Thesis: EWU Only

Discipline thesis:degree_discipline

Computer Science and Electrical Engineering

Year

2023

Author dc:creator

• Henderson, Michael J.

• Other Computer Sciences

Statement dc:rights

• Access perpetually restricted to EWU users with an active EWU NetID

Repository record dc:identifier

https://dc.ewu.edu/theses/863

OAI identifier oai:identifier

oai:dc.ewu.edu:theses-1861