For more than twenty years, VHS tapes were the main storage medium for storing recordings. The magnetic tape on which the recording is stored is quite unreliable and therefore a target for various degradations that can ruin the experience of watching the recording. To remove these degradations, we enhanced the deep learning method TAPE, which is based on a transformer architecture. We trained the model on authentic VHS recordings obtained by loading the recordings onto VHS tape and digitizing them from it. We learned the general model and three specialized models, which we evaluated and compared. The best results were achieved by the general model which improved evaluation metrics by 20 to 30 %.
|
