12.01.2024, 10:56
I have been researching about restoring analog tapes and getting the most out of them. This came up and I had to share. Not sure if it can be implemented in any way in hybrid. This study artificially added standard analog errors and trained the model to clean up those errors. This includes tracking errors, dropouts, chroma errors, temporal errors (wavy video). Really amazing! This model beats out everything I have seen and the data in the paper backs it up. I wish I knew a how to set it up to run some test footage.
Link to research paper: https://arxiv.org/abs/2310.14926
Link to github: https://github.com/miccunifi/TAPE
Link to research paper: https://arxiv.org/abs/2310.14926
Link to github: https://github.com/miccunifi/TAPE