A high-quality frame rate up-conversion technique for Super SloMo

摘要

In this paper, we propose several methods to improve Super SloMo, a deep learning-based frame rate up-conversion technique for the temporal quality improvement of video. In the proposed methods, the training dataset and hyper-parameter are changed and trained to obtain optimal results while maintaining the existing network structure of Super SloMo. The first method improves the cognition of images when trained with the validation set of characteristics similar to the training set. The second method reduces video loss in all validation sets when trained by adjusting the hyper-parameters of the error function value. The experimental results show that the two proposed methods improved the peak signal-to-noise ratio and the mean of the structural similarity index by 0.11 dB and 0.033% with the specialised training set and by 0.37 dB and 0.077% via adjusting the reconstruction and warping loss parameters, respectively.